Stories by Were Samson Bruno on Medium

Prompt engineering frameworks to get great results with AI IDEs

<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KYa5C2dmi7QdB59inZr6Jg.png"></figure><h3>My bloat wars with Windows Update.</h3>
There’s a peculiar frustration that comes with being a Windows Insider Preview participant. You sign up with the best intentions, excited to test cutting-edge features and contribute to Microsoft’s development process. Then reality sets in. Your computer starts downloading updates you never requested, updates that consume precious storage space, updates that simply won’t stop coming. Even when you disable automatic updates, thinking you’ve finally won the battle, Windows gives you only a temporary reprieve of one week before the cycle begins again.
This morning, I discovered something that changed everything. After weeks of watching my available storage shrink and feeling increasingly powerless against the relentless tide of updates, I found a solution that freed up 4GB of space on my machine. The answer wasn’t hiding in some obscure third-party application or requiring a complete system reinstall. It was right there in the Command Prompt, waiting for someone desperate enough to go looking.
<h3>The Problem That Keeps Growing</h3>
Let me paint you a picture of what led me here. As a Windows Insider, my computer receives a constant stream of preview builds and updates. Microsoft, in their wisdom, has designed the system to be persistent about these installations. You might think that toggling off automatic updates in your settings would solve the problem, but you’d be mistaken. The system allows you to pause updates for exactly seven days, after which it cheerfully resumes downloading and installing whatever Microsoft decides you need.
The real frustration came when I tried to investigate where all my storage space had gone. I opened File Explorer, searched through my folders, looked in all the obvious places where large files might accumulate. Nothing stood out. The Settings app has an update history section, but try as I might, I couldn’t find any option to delete these accumulated updates. This seemed deliberate, a design choice that keeps users from managing their own systems.
That’s when I remembered the oldest and most reliable tool in any Windows power user’s arsenal: the Command Prompt. If the graphical interface won’t let you do something, the command line often will.
<h3>Understanding What’s Actually Happening</h3>
Before we dive into the solution, it helps to understand what Windows is doing behind the scenes. When your computer downloads updates, it doesn’t just grab a small file and install it immediately. The operating system creates an entire cache of downloaded update packages, installation files, and historical data in a folder called SoftwareDistribution. Think of this folder as a warehouse where Windows stores everything related to updates, both current and past.
Over time, especially for Insider Preview participants who receive frequent updates, this warehouse becomes cluttered with files that serve no practical purpose anymore. Completed updates leave behind their installation packages. Failed updates leave behind partial downloads. The system keeps historical records of everything it’s done. All of this sits on your hard drive, consuming space that could be used for your actual files and programs.
The beauty of what I discovered is that you can safely clear out this warehouse without harming your system. Windows doesn’t need these old files to function properly. It’s like cleaning out a storage unit filled with boxes from moves you made years ago. The memories might be there, but you don’t need to keep the boxes.
<h3>The Solution: Step by Step</h3>
Here’s how I reclaimed my storage space, explained in a way that shows not just what to do, but why each step matters.
The first thing you need to do is open Command Prompt with administrative privileges. This is crucial because the folders and services we’ll be working with are protected system components. Press the Windows key on your keyboard and type “cmd” into the search box. When Command Prompt appears in the results, don’t just click it. Right-click on it instead and select “Run as administrator.” Windows will ask you to confirm that you want to allow this program to make changes to your device. Click yes. This elevated access gives you the permission needed to modify system files and control Windows services.
Now comes the interesting part. Windows Update isn’t just one program running in isolation. It’s actually a collection of services working together. Three services in particular handle the downloading, installing, and cataloging of updates. Before you can safely delete any update files, you need to stop these services. It’s like pausing a machine before you open it up to clean the parts inside.
Type the following command and press Enter:
<pre>net stop wuauserv</pre>
This command stops the Windows Update service itself, the primary component responsible for checking for and installing updates. You’ll see a message telling you the service was stopped successfully. Now type this command and press Enter:
<pre>net stop bits</pre>
BITS stands for Background Intelligent Transfer Service, and it’s the component that actually downloads update files in the background while you’re working on other things. Finally, type this command and press Enter:
<pre>net stop cryptsvc</pre>
The Cryptographic Services component validates and catalogs the updates to ensure they’re legitimate and haven’t been tampered with. With all three services stopped, you’ve essentially put Windows Update into a safe, dormant state.
Now you’re ready to navigate to where all those update files are hiding. Type this command:
<pre>cd %SystemRoot%\SoftwareDistribution</pre>
This command changes your current directory to the SoftwareDistribution folder. The %SystemRoot% part is a variable that automatically points to wherever Windows is installed on your system, typically C:\Windows. You’re now standing at the doorway of that cluttered warehouse I mentioned earlier.
At this point, you have two options depending on how thorough you want to be. If you want to delete just the downloaded update files while preserving some of the update history and metadata, type this command:
<pre>del /f /s /q Download</pre>
This command deletes everything in the Download subfolder. The flags tell Windows to force deletion of read-only files, search through all subdirectories, and operate in quiet mode without asking for confirmation on every file. If you want to go further and clear out the entire update cache and history, use this command instead:
<pre>del /f /s /q *.*</pre>
This removes everything in the Software Distribution folder. Note that I’m not telling you to delete the folder itself, just its contents. The folder needs to remain because Windows Update will recreate the necessary structure when it starts up again.
After the deletion completes, you need to restart those three services you stopped earlier. Type these commands, pressing Enter after each one:
<pre>net start wuauserv net start bits net start cryptsvc</pre>
You’re essentially turning the machine back on after cleaning it. Windows Update will now start fresh, recreating only the files it needs for future operations. Close the Command Prompt window and restart your computer to ensure everything initializes properly.
<h3>The Results and What I Learned</h3>
When I followed these steps this morning, I watched as Windows deleted thousands of files from the Software Distribution folder. The entire process took less than five minutes, and when it finished, I had reclaimed 4GB of storage space. That might not sound like much if you have a spacious hard drive, but on a laptop with limited SSD capacity, it made a real difference.
More importantly, I learned something valuable about Windows and about computing in general. The graphical interfaces we use every day are designed to be user-friendly, but they’re also designed to prevent users from doing things Microsoft has decided they shouldn’t do. Sometimes those decisions make sense for security or stability reasons. Other times, they’re just frustrating limitations that prevent you from managing your own computer.
The Command Prompt, that black window with white text that intimidates so many users, is really just a more honest interface to your operating system. It doesn’t hide options or make decisions for you. It simply does what you tell it to do, assuming you have the proper permissions. Learning to use it, even just for simple tasks like this one, gives you back a degree of control over your machine.
If you’re a Windows Insider like me, or if you’re just tired of watching Windows Update consume your storage space without your consent, I encourage you to try this process. Be careful, follow the steps exactly as written, and don’t delete anything beyond what I’ve described here. But don’t be afraid of the Command Prompt either. It’s been part of Windows since the very beginning, and it’s one of the most powerful tools you have for maintaining your system.
Your computer should work for you, not against you. Sometimes you just need to know where to look to make that happen.
Have you dealt with Windows Update frustrations? Found your own solutions to reclaim storage space? I’d love to hear about your experiences in the comments.
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=296ae4f6be65" width="1" height="1" alt="">

How to free up space on windows using CMD

<h3>How do we build truly “democratic” AI for Africa ? — My thoughts</h3>
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fdftcs8B6oN32JqPmqxFZQ.png"><figcaption>An AI generated image from Gemini with prompt “Generate an image for a medium article titled How do we build truly “democratic” AI for Africa with Africa as a priority? — My thoughts”</figcaption></figure><h3>Reflections from Norrsken Africa/Week 2025</h3>
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1fAZ_UqCH-bCS17CEtFapQ.jpeg"><figcaption>Norrsken House Kigali main entrance</figcaption></figure>From November 20th-21st 2025; I was attending the Norrsken Africa/Week in Kigali at the Norrsken House. I met, spoke to and listened to brilliant minds; great founders with powerful insights, perspectives and histories that have endured a lot of hardship so they can build solutions to not only African but global problems too.
The event brought together over 1,200 community members, partners, speakers, and attendees — a testament to the growing momentum of Africa’s innovation ecosystem. What struck me most wasn’t just the scale of the gathering, but the depth of the conversations happening in the corridors, panel discussions, and spontaneous encounters throughout Norrsken House Kigali.
<h3>Common Sentiments: Data Sovereignty, Talent, and Opportunity as Infrastructure</h3>
Some of the commonest sentiments I have heard and resonated with me is data sovereignty, building talent as an infrastructure but also providing avenues/opportunities for this talent to be able to shine and play a pivotal part in building and contributing to the AI revolution for Africa.
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_TBA5zGTzqh2MFy8mOTJQw.jpeg"><figcaption>Mathieu (L) and Mark (R)</figcaption></figure>This I gleaned from the session moderated by <a href="https://medium.com/u/b1a31c9b97dd">Mathieu AGUESSE</a>, Co-CEO, Schoolab and one of the panellists <a href="https://medium.com/u/4c7dfa14fc00">Muchiru Mark Kuria</a>; Principal at Gaia Impact was speaking on Corporate Capital &amp; African Innovation: Funding the Next Decade; and while he did he hinted on how he has been and is actively thinking about building Data centres for Africa that are powered and run by renewable energy sources like geothermal, solar, among others and this piqued my interest because this is something most of my colleagues and friends hear me “lecture/rant” on or about. I am big on Africa owning it’s data and having the rights to control how that data is handled; now more than ever it’s important with the AI age/revolution on it’s cusp.
These themes weren’t abstract talking points — they are urgent questions we need to ask ourselves as founders, investors, ecosystem enablers, governments and well-wishers.
<h3>Cassava Technologies: A Case Study</h3>
One of the companies that is championing data sovereignty in Africa is Cassava Technologies. They provide internet infrastructure in through one of their subsidiary companies <a href="https://liquid.tech/">Liquid Intelligent Technologies</a>
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*hogMpL9fV6FMmLW8.jpg"></figure>which offers a range of services as listed here:- - Extensive Fibre Network: The company has laid over 110,000 km of fibre optic cable, creating Africa’s largest independent fibre network, including a link stretching from Cape Town to Cairo.
<ul>
<li>
Cloud Services: Liquid partners with global leaders like Microsoft and Google Cloud to offer cloud solutions, including deploying Azure Stack in countries such as Uganda, Kenya, Zambia, and Tanzania to meet local data regulatory requirements and reduce latency.</li>
<li>
Data Centres: They operate a network of interconnected, carrier, and cloud-neutral data centres in major cities like Johannesburg, Cape Town, Nairobi, Harare, and Kigali.</li>
<li>
Cyber Security: The company provides a range of cyber security solutions and has launched AI-powered programs and a Cyber Security Business Unit.</li>
<li>
Connectivity Expansion: Liquid is actively working on new fibre routes and leveraging subsea cables to enhance high-speed connectivity across the continent, including previously underserved areas.</li>
<li>
Digital Transformation Partnerships: They collaborate with governments and local businesses to drive digital transformation, such as providing digital computer centres in South Africa and enabling remote monitoring and data analytics for the mining industry in Botswana.</li>
</ul>
I have interfaced with their services in Uganda, Rwanda, South Sudan and Kenya but their influence and network spans across 18 African countries; from the cape to the heart of Africa in Cairo. This is by far the largest provider of Internet Infrastructure for Corporate use cases.
They (Cassava Technologies) have another company that is called <a href="https://www.cassava.ai/">Cassava AI</a> and this is aimed at providing GenAI for every Telecommunications, Financial Services, Universities and Higher Education Institutions, Mining, Governments, Hospitality and Tourism, Agriculture and Manufacturing.
Cassava has partnered with Accenture and NVIDIA to deliver sovereign AI solutions, ensuring that African data stays within the continent and complies with local regulations. They’ve launched <a href="https://www.cassavatechnologies.com/cassava-technologies-launches-africas-first-ai-multi-model-exchange-to-advance-local-ai-development/">CAIMEx</a>, Africa’s first AI Multi-Model Exchange, making advanced AI models accessible to mobile network operators across Africa. Their infrastructure spans over 110,000km of fibre network and multiple data centers across the continent.
On paper, this sounds transformative but what is the common man on the ground benefiting from this? How does the regular African tap into this and benefit from it?
<h3>Are the masses in the room?</h3>
Providing AI for these sectors is great however what is the pathway to providing the same services to entrepreneurs or the general masses? Cassava AI’s route seems more enterprise and corporate than it is general or consumer targeted. This is something I believe needs to change.
While Cassava Technologies builds the highways of African AI infrastructure, most entrepreneurs and everyday users are still searching for the on-ramp. The enterprise focus makes business sense — these are the clients with resources to pay for GPU computing power, data centers, and AI integration services. But this approach risks creating a two-tier system where AI remains a tool of large institutions rather than a force available for everyone. AI like Internet has become or is about to become a necessity and need; we can’t afford as a continent to discriminate or alienate the masses from accessing it.
<h3>The Big Question: How Do We Democratize AI?</h3>
I have had questions on what is the right way to democratize (make AI in general — not only generative) accessible to and usable by the general masses that aren’t technical gurus/experts. What does this mean in terms of choosing the medium of communication? What even is democratization in this context?
This question becomes even more complex when we consider Africa’s linguistic landscape.
Majority of African languages are oral and not necessarily written down on tablets or as text. So how do we build for such a community, for such a people?
This isn’t just a technical challenge; it’s a fundamental question about what “AI accessibility” means. An English-language chatbot interface means nothing to a Kenyan farmer who speaks Kikuyu, a Ghanaian trader who conducts business in Twi, or a Rwandan entrepreneur whose first language is Kinyarwanda.
African languages account for at least one third of spoken languages worldwide, yet only a handful of products that offer AI to individuals support these languages — despite the fact that the majority of these languages exist in oral forms rather than in written text.
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qPBxP9Y5fMtazB8GGIhWQw.png"><figcaption>The web interface of sunflower web app</figcaption></figure>One of the heavy weights I know of that currently does this is recently launched <a href="https://sunflower.sunbird.ai/">Sunflower </a>Web app from <a href="https://sunbird.ai/">Sunbird AI.</a> The bar to be met is high and they are not yet there in terms of experience but it’s an effort worth noting given that they are a Non-profit and Sunflower is provided free of charge to use.
Of course there are other builders on the continent that I haven’t laboured to mention in this article for fear of making it very lengthy, but atleast we have vanguards in every corner of the continent building towards this African first AI. How they arrive at it and approach the problem might be different but I am rooting for all of them.
<h3>The Infrastructure We Need</h3>
I think if we got the the answers to these questions it’s the beginning of having a wonderful understanding of how to approach AI as the African continent but also how to build infrastructure and systems that support entrepreneurs building AI and building with AI.
The infrastructure gap isn’t just about fibre optic cables and data centres — though those matter immensely. We need:
<ol>
<li>
Voice-First AI Systems: Since many African languages are primarily oral, voice should be the primary interface, not text. This requires massive investments in speech recognition and synthesis for hundreds of languages.</li>
<li>
Accessible Compute Resources: GPU-as-a-Service is excellent, but pricing models need to accommodate entrepreneurs working on shoestring budgets, not just corporations with IT departments. This also boils down to charging entrepreneurs in local currency not USD/GBP.</li>
<li>
Localized Training Data: AI models need to be trained on African contexts, idioms, and use cases — not retrofitted from Western datasets. There’s a vast trove of information on African folklore that is stored up in elder’s minds that has yet to be digitized and electronically stored. Telecoms in Africa have access to an immense dataset of caller data which with the right compensation and appropriate data handling procedures could be turned into invaluable datasets to train AI that’s Africa first and for good. Consumer data is not just for Quality Assessment or Surveillance by the state and government as seen in <a href="https://www.business-humanrights.org/en/latest-news/kenya-safaricom-under-fire-as-senators-demand-answers-on-alleged-data-breaches/">Kenya </a>and <a href="https://www.privacyinternational.org/sites/default/files/2017-12/Uganda_Report_1.pdf">Uganda</a>
</li>
<li>
Developer Education and Support: Infrastructure means nothing without the human capacity to build on it. We need widespread AI literacy programs. This is to enable and build AI talent as an infrastructure for the continent’s use and for global use too. <a href="https://www.index.dev/blog/top-countries-ai-developer-talent-pools">United States is currently leading</a> in this regard with China and India closely following and I am certain we can join the race too if this is done early enough.</li>
</ol>
No African country is to be found anywhere on the <a href="https://www.forbes.com/sites/johnkoetsier/2025/09/11/top-10-ai-nations-global-ai-superpowers-ranked/">worlds top 10 AI nations</a>; nor on the <a href="https://hai.stanford.edu/news/global-ai-power-rankings-stanford-hai-tool-ranks-36-countries-in-ai">global AI power rankings</a> this should be a wakeup call for all our decision makers and folks with power.
<h3>Mozilla Common Voice: A Blueprint for Collaboration</h3>
<a href="https://commonvoice.mozilla.org/en">Mozilla Common Voice</a> provides a platform for individuals to be able to contribute to an opensource voice dataset of almost all African languages but it’s adoption isn’t widely spread; this is another avenue or initiative we can use to index, record and catalogue African voice data which can be accessed and used later on in subsequent waves of revolution(s).
Since 2019, Mozilla has been working with partners to promote the creation and use of open voice data in East African languages like Kinyarwanda, Kiswahili, and Luganda. The platform has grown to include 100 languages, with significant African representation. The <a href="https://datacollective.mozillafoundation.org/datasets/cmflnuzw6p5iadfhltsgb0vzo">Luganda dataset</a> is one of the datasets I contributed to and you too can become a voice of change (literally use your voice) <a href="https://commonvoice.mozilla.org/lg/speak">here </a>. You can speak, listen to correct or write.
But here’s the challenge: despite its potential, adoption remains limited. Why? Awareness is low, contribution requires internet access and devices that many don’t have, and the value proposition for individual contributors isn’t always clear. There’s a saying in Kampala “Nfunilamu wa?” loosely translated to “Where or how do I benefit from this?”
Yet this model — community-driven, open-source, focused on linguistic diversity — represents exactly the kind of infrastructure Africa needs. Imagine if contributing to voice datasets became as common as posting statuses on whatsapp or making a tweet or contributing to Wikipedia, with clear pathways from data collection to useful applications that serve contributors’ communities.
<h3>What can you do?</h3>
The gap between enterprise AI and democratized AI isn’t inevitable. Bridging it requires intentional design:
For Companies Like Cassava: Consider tiered pricing models, startup incubation programs, and partnerships with educational institutions. Make GPUaaS accessible not just to telecommunications giants but to the developer building a healthcare chatbot in a Nairobi co-working space.
For Policymakers: Invest in digital literacy programs that specifically address AI. Create grants and subsidies that enable small-scale entrepreneurs to access AI infrastructure. Support open-source voice dataset initiatives. Support and create policies that invite investment to the continent and do the basic stuff that won’t scare them away.
For Developers and Entrepreneurs: Contribute to projects like Mozilla Common Voice. Build applications that demonstrate AI’s value in local languages for local problems. Share knowledge and mentor others.
For Investors: Look beyond enterprise-scale solutions. Fund entrepreneurs working on consumer-facing AI applications, especially those addressing the needs of underserved communities. It’s not my place to say this but I will regardless. You need to grow a large risk appetite and have deep seated belief as deep as your doubt for the continent is; to fund entrepreneurs building crazy stuff on the continent; especially if it’s infrastructure or rails level and there’s no clear path to exit for your investment.
<h3>Bottomline: Africa’s AI Future Must Be Inclusive</h3>
The conversations at Norrsken Africa/Week made one thing abundantly clear: Africa is not content to be merely a consumer of AI technology. The opportunity I had to see real African movers and shakers made me understand the importance of African ownership in designing systems to fund African innovation, from basic sectors to emerging technologies like AI.
But ownership isn’t just about corporate control or data sovereignty — crucial as those are. True ownership means ensuring that every African entrepreneur, every developer, every innovator has the tools, infrastructure, and opportunity to participate in building Africa’s AI future.
The infrastructure being built by companies like Cassava Technologies is essential. But it’s only the foundation. We must build the bridges, the pathways, the on-ramps that connect this infrastructure to the millions of Africans whose problems AI could help solve, whose languages deserve to be spoken by machines, whose innovations could reshape not just Africa but the world.
The AI revolution shouldn’t have a VIP section. If we’re building for Africa, we must build for all of Africa.
What are your thoughts on democratizing AI in Africa? How can we ensure that AI infrastructure serves not just enterprises but everyday users and entrepreneurs? Comment to get the conversation started or reach out to me on <a href="https://www.linkedin.com/in/were-samson/">Linkedin</a>.
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=c5a1201730c8" width="1" height="1" alt="">

How do we build truly “democratic” AI for Africa with Africa as a priority? — My thoughts

We built and launched Intelli’s landing page expecting instant traction and ranking first on google searches because we used NextJs ; Duh— and got crickets. No Google love, organic clicks were there but mostly redirecting from socials; from here(thanks to the people who read my articles/stories), but when it came to Google Organic search the story was different. It was bewildering. We had built an AI powered multi-channel customer engagement and support platform (with all the bells and whistles), yet searchers couldn’t find us. For a couple of reasons I will list here in the preamble. 1. the Intelli domain we wanted and was available was way too expensive a spend for us at this stage (given our priorities) so we went for an affordable one <a href="http://www.intelliconcierge.com/">www.intelliconcierge.com</a> long but kind of aligns with what we were doing when we started out (Travel CS) 2. We had other players in the market who came before us. Hello Jetbrains with IntelliJ IDEA… You have shadowed us for far too long. So we asked ourselves a lot of questions on how to improve our SEO. Why wasn’t anyone finding us first? Why won’t we show up on Google’s First Page? Why won’t our page rank in the first 5 searches? If these are questions you have asked yourself and either didn’t find the asnwers or did but were not satisfied; this article is for you. Many founders skip SEO in the excitement of launch, then wonder why “Publish” didn’t spark a traffic fireworks show. We learned the hard way that SEO isn’t magic, but it is make-or-break for being found first on the internet.
In this post I’ll walk you through exactly what we did — step by step — to fix our SEO from the ground up. Whether you’re on Next.js, plain HTML, WordPress or anything in between, these tweaks apply. We overhauled our metadata and headings, sharpened keyword usage, clarified our site name, added structured data, and set up sitemaps/robots.txt. Best of all, none of it required hiring an SEO guru — just some elbow grease and free tools.
<h3>The SEO Reality Check</h3>
First, we confessed we had an SEO problem. Searching for our site using the keywords “Intelli” were very humbling. We were on the 2nd or 3rd page; Only our LinkedIn page showed up on the first page and this is because of two reasons; we post often and LinkedIn is crawled often for google searches. Logging into Google Search Console was very telling: it showed very negligible impressions for our main keywords, and the few clicks we had were from direct links shared or from social media. Some were purely accidental. As Google puts it, GSC shows “which queries bring users to your site. Analyze your site’s impressions, clicks, and position on Google Search”. In our case, the answer was “second to none,” because very few individuals knew we existed; according to our search. We had to improve this.
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nSKCq4bJvPEBKx9QbGAqaQ.png"><figcaption>A comparison of our performance in the last 28 days over the last year;</figcaption></figure>Next, we audited why. We realized our site was missing all the basic signposts that Google needs. Our &lt;title&gt; tags were generic, we had no or duplicate headings, and we hadn’t told Google about our pages (no sitemap!). In short, we’d built a gorgeous site but never optimized it for search engines. So we set out on an SEO makeover.
Here’s what we changed, and how you can do it on any stack:
<h3>Metadata Makeover: Title Tags &amp; Meta Descriptions</h3>
Your &lt;title&gt; tag and meta description are like the book cover and blurb for Google. We fixed ours by making titles descriptive, keyword-rich, and brand-inclusive. For example, we changed our home page title from
<pre>&lt;title&gt;Intelli&lt;/title&gt; (We changed this from Intelliconcierge because we wanted to rank for Intelli)</pre>
to something more like:
<pre>&lt;title&gt;Intelli – Engage and support customers &lt;/title&gt; "Add a short snippet of what your platform does."</pre>
This new title includes our main keywords (“Engage and support Customers ”, “Intelli”) and our brand. We kept it under ~60 characters to avoid truncation. (Pro tip: Google typically shows about 65–70 characters or ~580–600px of a title.)
We also rewrote our meta description to be a concise, inviting summary of the page with a call to action, around 150–160 characters. For example: “Intelli helps businesses automate customer service using AI. Sign up now for improved customer support!” Google usually shows up to ~960px of meta description on desktop, so roughly 150–160 chars. We used a SERP snippet preview tool to test these lengths.
These changes alone turned our search snippet from vague to clickable. A clear, keyword-rich title catches attention, and a helpful description boosts click-through rate (CTR). Remember, Google uses your meta tags to generate the search snippet. Keep each page’s title and description unique and avoid stuffing in too many keywords. Instead, sprinkle in one or two main terms naturally.
Golden Nuggets:
<ul>
<li>
Title tag: Include primary keyword + brand, keep around 50–60 characters.</li>
<li>
Meta description: Write a helpful summary with at least one keyword, ~150–160 characters.</li>
<li>
One &lt;h1&gt; per page: Make it match your title theme (more on headings below).</li>
<li>
Alt text: Don’t forget image alt attributes – a few words describing each image helps image search.</li>
</ul>
<h3>Content &amp; Structure: Headings and Keywords</h3>
Once our metadata was in order, we turned to page content. Search engines love well-structured pages. We audited our content and headings like nit-picks: making sure we had one &lt;h1&gt; as the page title, then logical &lt;h2&gt;–&lt;h3&gt; sections. We updated our homepage to have an H1 that mirrors the title tag (e.g. “Intelli — Engage and Support Customers”), so Google sees a consistent topic. Each section under that got meaningful subheadings.
Next, we sharpened our keyword usage. We researched the terms people actually search. Tools like <a href="https://trends.google.com/trends/">Google Trends</a>, <a href="https://business.google.com/in/ad-tools/keyword-planner/">Keyword Planner</a>, and even the “Performance” report in Search Console helped. We asked, “What would someone type if they needed a platform for AI support to engage with their customers and warm leads?” Likely queries included “AI multi-channel software,” “automated customer support,” or “Engage customers” “Marketing to customers”. We made sure these phrases appeared naturally in our content, especially in the first paragraph and headings. We didn’t cram keywords unnaturally — Google frowns on that — but used them in context. For example: “Intelli is an AI-powered customer support multi-channel platform that helps businesses engage customers and leads.”
If you have multiple pages, avoid competing with yourself on the same keywords (this is called “cannibalization”). Check Search Console to see if two pages vie for a single term, then merge or distinguish them.
We also took stock of our content’s readability. Short paragraphs, bullet points, and clear language not only help readers but help SEO. Google’s algorithms notice when users engage longer with the content.
<ul>
<li>We added a couple of bullet lists (like this one!) to break up text. For example, we listed Intelli’s key features in bullets, which made it easier for users and gave Google clear signals of what’s important on the page.</li>
<li>A clear way for your business might be answering a question they have in a straightforward way for example; How much does [insert company name here] charge per month? — We charge [insert amount you charge] in USD or whatever currency. Google picks up such responses faster and improves your ranking on Google search.</li>
</ul>
<h3>Structured Data: Speaking Google’s Language</h3>
To make our site even more search-engine-friendly, we added <a href="https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data">structured data (schema)</a>. This is just a fancy way of saying we gave Google explicit hints about our content using JSON-LD code. On our homepage we implemented the “Organization” schema for Intelli Concierge: name, logo, social profiles, and so on. We also added a BreadcrumbList schema so Google could understand our navigation. These snippets help generate rich results (like knowledge panels or site links).
Google Search Central encourages adding as many relevant properties as apply; they say some fields help “disambiguate your organization” and even influence which logo shows in search results. In practice, that means including our company name, URL, logo URL, and contact info in the JSON-LD script. We pasted something like:
<pre>&lt;script type="application/ld+json"&gt;{ "@context": "https://schema.org", "@type": "Organization", "name": "Intelli", "url": "https://www.intelliconcierge.com", "logo": "https://www.intelliconcierge.com/logo.png", "sameAs": [ "https://twitter.com/intelli", "https://linkedin.com/company/intelli-concierge" ] }&lt;/script&gt;</pre>
Even though structured data doesn’t directly boost rankings, Google’s docs show that pages with schema saw up to 25–82% higher click-through rates in case studies. In short: we made our site understandable, and Google rewarded us with better search features.
Our Structured Data Tips:
<ul>
<li>Use a JSON-LD schema generator or CMS plugin to add schema (if you’re on WordPress, plugins like Yoast can help).</li>
<li>On static/Next.js sites, paste a &lt;script type="application/ld+json"&gt; block in your HTML template.</li>
<li>Include Organization schema on your homepage (name, logo, social, etc.).</li>
<li>If you have articles, add Article schema (title, author, date). If you list products or services, add Product or Service schema.</li>
<li>You can test with <a href="https://search.google.com/test/rich-results">Rich Results Test</a> and fix any flagged issues.</li>
</ul>
<h3>Sitemap &amp; Robots.txt: Guiding the Crawlers</h3>
Next, we made sure Google could crawl our site properly. We generated an XML sitemap listing all our important pages (You can use ChatGPT and a ton of other free tools online, or plugins if you use a CMS). Then we submitted that sitemap URL in Google Search Console, under Index &gt; Sitemaps. We also added a line in our robots.txt file pointing to the sitemap (e.g. Sitemap: <a href="https://www.intelliconcierge.com/sitemap.ts)">https://www.intelliconcierge.com/sitemap.ts)</a> for your case it will be sitemap.xml if you are not using NextJs; so crawlers can find it automatically.
Why bother? Google explicitly recommends sitemaps: “A sitemap helps search engines discover URLs on your site… in most cases, your site will benefit from having a sitemap”. It doesn’t guarantee crawling, but it gives the crawler a roadmap. After submitting ours, Search Console started showing index coverage data, telling us how many pages were indexed or if any were blocked.
Speaking of blocking: we audited robots.txt carefully. We only disallowed obviously private or redundant paths (for example, /api/* endpoints or temporary calendar links). We didn’t block any of our main pages. Remember, Google’s guide warns that robots.txt is mainly to manage crawler traffic, not to hide content. In fact, if you block a page in robots.txt, Google might index it without reading content! So use noindex meta tags on pages you want hidden, not robots.txt (which should almost always allow your public pages). In our case, we ensured the robots.txt looked like:
<pre>User-agent: * Allow: / Disallow: /api/ Disallow: /private/ Sitemap: https://www.intelliconcierge.com/sitemap.xml</pre>
This way, crawlers could easily access everything we wanted them to see.
<h3>Keeping Tabs: Tools &amp; Testing</h3>
Throughout this process we leaned heavily on tools — mostly free — to track progress. Google Search Console was our north star: every week we checked the Performance report for impressions and clicks by query and page. GSC truly “helps measure your site’s search traffic and performance… fix issues, and make your site shine in Google Search results”. When our changes took effect, we started seeing impressions rise for our target keywords, and our average position improved.
We also used:
<ul>
<li>
Google Analytics: to verify that organic search visits started ticking up and check bounce rates.</li>
<li>
Keyword research tools: Google’s own Keyword Planner (free with a Google Ads account) and Google Trends helped confirm that we were using terms people actually search for.</li>
<li>
PageSpeed &amp; Core Web Vitals: Since load speed is now a Google ranking factor, we ran Lighthouse audits and the Web Vitals report in Search Console. Google recommends monitoring Core Web Vitals (LCP, FID, CLS) on mobile and desktop. We optimized images and enabled caching to get green scores.</li>
<li>Another tool I had almost forgotten to add was <a href="https://ahrefs.com/">Ahrefs</a>; I discovered this for backlinks and search ranking was overall most accurate.</li>
</ul>
Quick tool tips:
<ul>
<li>In Search Console Performance, filter by page or query to see if your tweaks are working (new clicks = good!).</li>
<li>Use the URL Inspection tool in Search Console to “Request Indexing” when you publish or update a page. It’s a quick way to tell Google to re-crawl that page.</li>
<li>Try Screaming Frog (free version) or a similar crawler to audit on-page SEO (it will show missing titles, broken links, duplicate content, etc.).</li>
<li>If you want ranked insights, tools like Ahrefs or SEMrush offer free trials and can spy on competitors’ top keywords. But you don’t absolutely need them — start with Google’s free reports.</li>
</ul>
<h3>Results &amp; Lessons Learned</h3>
After implementing all these changes, our SEO performance turned a corner. Within a month we saw the first glimmers of traffic: impressions went from near-zero to hundreds per day, and clicks began appearing for our target phrases. It’s hard to quantify exactly how much each change helped, but the combo of better titles, headings, and schema definitely improved our visibility. More importantly, our click-through rate (CTR) jumped because our search listings finally looked relevant and trustworthy. What was once a sad “0” in Search Console started to climb.
The key lesson? SEO is a long game of fundamentals. You won’t rocket to page one overnight, but you will rank higher (and attract real users) once you do the basic housekeeping. Every site — no matter the tech stack — should have well-crafted metadata, logical headings, a clear site structure, and a sitemap. Structured data and robots.txt are icing on the cake. These steps are mostly free and under your control; it just takes time and attention to detail.
If you feel overwhelmed, pick one thing and do it: start with the title and description on your homepage. Then audit Search Console for obvious issues (errors or excluded pages). Step by step, you’ll build momentum. SEO isn’t some mystical secret — it’s about making your site understandable and appealing, both to Google and to human visitors.
<h3>As I wrap up to get out of here.(TLDR)</h3>
Ensure to remember these:
<ul>
<li>
Optimize Metadata: Catchy title + meta desc with keywords (watch character limits).</li>
<li>
Structure Content: One &lt;h1&gt; per page, strategic &lt;h2&gt;s, and naturally include your main keywords.</li>
<li>
Add Schema: Use JSON-LD for Organization, Articles, Products, etc., to enable rich results.</li>
<li>
Sitemap &amp; Robots: Submit an XML sitemap (Google loves it) and make sure robots.txt isn’t blocking anything important.</li>
<li>
Monitor &amp; Iterate: Use Google Search Console religiously to track queries, clicks, and fix any issues that crop up.</li>
</ul>
By doing this, you’ll move from obscurity to being visible. And being visible means more potential customers discovering you daily.
If you’d like to see these principles in action, check out <a href="https://www.intelliconcierge.com/">Intelli </a>— our own AI customer support platform. We believe in practicing what we preach (yes, we walk the talk !), so we apply these SEO tactics to our product site too. Give it a browse and see if our improved search listing stands out.
Ready to give your site an SEO tune-up? Start with your homepage metadata and a quick run through Search Console. Then pick one more area (headings, content, or schema) and fix it. Before long, Google will start rewarding your efforts. And who knows — maybe your site will be the next case study inspiring other founders/startups especially your competitors with big budgets to spend.
References: We followed Google’s official guidance and SEO best practices throughout this process. These changes are grounded in what’s proven to work in 2025 and beyond. Have a wonderful day and may your star shine brighter as you are noticed by the world, and may your search rankings soar! Adios
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=cbdca584e0ef" width="1" height="1" alt="">

How We Rebuilt Intelli’s SEO And What You Can Learn From It

<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LUQ0F1SL2V2i40nRp0CbPg.jpeg"><figcaption>An AI generated image From Meta AI with prompt "Generate an image for a medium article titled: principles over time learn</figcaption></figure>When I looked back on the past 12 months we’ve built <a href="http://intelliconcierge.com/">Intelli</a> for.
It made a lot of sense why I had to have a support system with me. And a support system isn't necessarily the people(individuals who promise and tell you how they are available and there for you) the real support system is made up of folks who show up consistently on days when situations demand. These people don't peddle words or promises. They show up with fervrour and do what needs to be done.
We received funding from an accelerator. That money(investment decision was made in September 2024) and fast forward we chose a path of incorporation that had us wait for a good solid 5 months( we received our incorporation certificate in January 2025) after that the battle for the EIN began. If we had a US citizen as a cofounder these are things that could have been done instantly: talk of the power of a SSN. I have been reminded profoundly of how different worldly systems treat anyone who’s not of their system or one considered to be from a lower grade of a country.
Like a colleague always says "We are treated like second class citizens". While I'm not one to ascribe to this train of thought or to think of myself as a second class (or any class) citizen for I come from above, I certainly have felt the impact of this thinking and notion in policy.
My mandate today isn’t to whine but rather to emphasise and impart key lessons that have been very important and crucial to keeping me together throughout these tribulations and challenges that shouldn’t have occured in the first place.
<h3>Deter you’ll not incur any costs:</h3>
This is the same principle used by the USA in war, whenever you spot a potential threat or problem that might arise from taking a certain path or making a certain decision, it’s best to avert that from happening by outright taking another route or measures that prevent what you anticipate to occur from happening.
By so doing you do what's called nipping a problem in the bud. For our instance our incorporation should have come through way quicker if we had used a platform and structure which was already established and well trusted. This would enable us to move smoothly through creating bank accounts and all related processes post incorporation wouldn't be so painful to us and affect our plans and operations. We would have saved 6+ months burnt away moving and doing business like a self funded startup yet we are VC backed and the money ought to give us the time advantage.
<h3>It’s a marathon not a race:</h3>
Plan wisely and move strategically. When you observe Usain bolt or any 100M race sprinter run and at the same time compare their style, speed, cadence and rhythm with a long distance runner like Eliud Kipchoge who does marathons of 2500~5000M you learn the difference between calculated energy expenditure.
A startup typically takes 10-15 years to take off the ground and become something, that's if it survived the first 3 years of infancy. For us it's not yet even the 1 year mark so we have a long way to go. This thinking has been ingrained in me and I constantly are awakened to this consciousness to be a little bit more patient with processes and how they pan out because in the grand scheme of things the macro picture looks elegant and grand. The pain and suffering experienced in the 5-6 months is nothing compared to the eternal glory to be enjoyed when we turn 15, 20, 40, 100 etc.
So how we plan our marketing campaigns, branding, product development, and delivery is all with this marathon mindset. If we don't die in the next 10 years we are going to build a unicorn, and even if we die we shall have built a framework and system that can survive the impact of time itself.
<h3>No one is coming to save you:</h3>
There’s nothing that is sobering to the mind like knowing that no help is coming from outside. All the help you will ever need is inside you. You just need to dig deeper to find it.
Knowing that we we're stranded on a startup island in rough seas of bills from ISPs, server hosting companies and many other things kind of sobered me up to start looking for solutions within and top looking outward( to the portfolio and our PoC) This triggered a mindset of there's no impossibility within me, as long as I can think of it, I'll do it. I'll need to take time to study hard and burn the midnight oil but when I pay the price in full I'll finally get to do what I set my mind to. Our product has grown tremendously and so has the team grown with it too. We are all better Software Engineers than we were a year ago. And that's a testament to knowing that there's no outside help coming. Ofcourse this doesn't actually mean that you cut off everyone. If they do come through and genuinely offer help do accept it humbly and keep it moving.
Keep it together your mind needs you in one piece. Your spirit man has a lot of answers but most of the times the processing part of you(your mind) is confounded either by external stimuli and thus cannot listen or heed to the answers from within or is unable to understand the language he(the inner man) uses and thus in computer science terms the mind cannot compile the machine language the spirit speaks into high level human readable programming languages that are easily understandable by and articulated to another man. So we have to switch off the external stimuli to sort the first problem and then learn by either trial and error or divine enablement to speak the language that the spirit speaks with our mind. This way we get answers to the hardest questions we've asked ourselves.
<h3>Do hard things first</h3>
Over the course of now 10 months I've been building our product. I haven't gotten a single ounce of pleasure out of staving off or delaying to do a hard thing (translated into something that requires a lot of brain power, focus and time investment)
Most of the times we get dopamine heights from one-shotting tasks, those tasks that get done so easily without even thinking so hard about it or lifting a finger. These tasks are good but unfortunately they don't add to our growth neither do they benefit us or our stakeholders in the long run.
It is only from doing hard things that I have come to learn a lot, I have also come to appreciate what I thought my limits were, surpassed them and as a result becomes better as a SWE.
It is while doing hard things that I got epiphanies on how to do other things simpler and in more straightforward ways because I spent time to think about how efficient a process could be made or jow much more lag we could squeeze out of an API call so the user has a smooth experience interacting with our product.
Those days when I look at my task board and it's a hard task I know greater days lie ahead of the constant persistent errors and poring through documentation and stack overflow thread to arguing with AI in my IDE.
Do hard things, for that's what greatness is called to do.
<h3>Travel lean/light, Go together.</h3>
Finally, this principle will keep you as light as possible; for anyone going on a marathon or a long distance flight travelling light/lean might seem counterintuitive since the journey is gonna be long and thus you need to prepare for it with rations and food and everything that you need.
However you'll be reinforced with the second part of the statement; go together. Hikers on a mountain trail can survive better when huddled together for warmth and protection from the elements and wild animals better than hikers who travel solo however much well prepared they are.
Having wonderful partners in your teammates is an essential block for you to achieve this. Make sure everyone has a piece of the rope they are holding such that if one of you slips the others can feel it and uphold them.
Doing this in turns will ensure your energy is preserved and will enable all of you to make it through the long tedious journey. Sometimes it's a simple wire transfer, a word of encouragement, a praise, a hug, a pat on the shoulder, a video call, and it'll go a long way to push your cofounder out of that corner the world has boxed them into back into the main game.
Going together also means no pointing fingers. It's very easy to shift blame (trust me it's very tempting) to find someone to blame for all the things that are not working as expected but kindly acknowledge the problems and find solutions for them rather than allocating blame to who or what is causing these problems. Ofcourse if ever in doubt about how to solve problems refer to principle 1. There you go folks.
<blockquote>That's what I've gleaned from my experience I hope it helps you too. Stay empirical, stay based, stay level headed. Have the faith of God.</blockquote>
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=aaea2549652d" width="1" height="1" alt="">

Principles over time learned: my lessons over the past 12 months

<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_aztuZXk9Y-hfTec2pc-CA.jpeg"><figcaption>Image generated from Gemini AI with prompt "Generate an image and make sure these words are visible: "how to restore a lost upload key 🗝️ for Google Play store" display a laptop with the article showing on the laptop screen and a well lit window in the background"</figcaption></figure>Losing your app’s upload key can feel like a major setback, but thankfully, Google Play provides a way to recover. This article guides you through the process of resetting your upload key, whether you use Android Studio or Visual Studio Code for development.
Understanding the Importance of Upload Keys
Before diving into the process, it’s crucial to understand the role of upload keys in the Google Play ecosystem. Your upload key is what you use to sign your app bundles (*.aab files) before uploading them to the Play Store.1 Google verifies this signature to ensure the app genuinely comes from you. Google then re-signs your app with their app signing key (which they manage) before distributing it to users.2 This two-step signing process enhances security.
Scenario: You’ve Lost/Compromised Your Upload Key
If you’ve lost your upload key (keystore file) or suspect it’s been compromised, you must generate a new one and request a key reset through the Google Play Console.3
Method 1: Using Android Studio (For Key Generation)
If you use Android Studio, you can generate a new upload key and keystore as follows:
<ol>
<li>
Open the “Generate Signed Bundle/APK” Dialog: In the menu bar, click Build &gt; Generate Signed Bundle/APK.4</li>
<li>
Select “Android App Bundle” or “APK”: Choose the appropriate option and click Next.</li>
<li>
Create a New Keystore: Click Create new below the “Key store path” field.</li>
<li>Fill in the Keystore Information:</li>
</ol>
<ul>
<li>
Key store path: Choose a secure location and filename (e.g., my_new_upload_keystore.jks). Remember this path!
</li>
<li>
Password: Create a strong password for your keystore. Remember this password!
</li>
</ul>
<ol><li>Fill in the Key Information:</li></ol>
<ul>
<li>
Alias: Enter a descriptive name for your key (e.g., upload_key). Remember this alias!
</li>
<li>
Password: Create a strong password for your key. It’s recommended to use the same password as your keystore password for simplicity, but it can be different. Remember this password!
</li>
<li>
Validity (years): Set this to at least 25 years.</li>
<li>
Certificate: Provide the requested information. This is not displayed in your app.</li>
</ul>
<ol><li>
Click OK: This generates your new keystore and key. Back up this .jks file immediately!
</li></ol>
Method 2: Using the Command Line (For Key Generation)
If you use Visual Studio Code or prefer the command line, you can generate the key using the keytool utility (included with the JDK):
<ol>
<li>
Open Your Terminal: Navigate to the directory where you want to store your keystore.</li>
<li>
Run the keytool Command:
</li>
</ol>
Bash
<pre>keytool -genkey -v -keystore &lt;your_keystore_name&gt;.jks -alias &lt;your_key_alias&gt; -keyalg RSA -keysize 2048 -validity 9125</pre>
<pre>* Replace `&lt;your_keystore_name&gt;.jks` with your desired filename (e.g., `my_new_upload_keystore.jks`). *Remember this path!* * Replace `&lt;your_key_alias&gt;` with a descriptive alias (e.g., `upload_key`). *Remember this alias!*</pre>
<ul>
<li>
Provide Information at the Prompts: You’ll be asked for keystore and key passwords (choose strong ones and remember them!) and certificate information.</li>
<li>
Verify Keystore Creation: Run keytool -list -v -keystore &lt;your_keystore_name&gt;.jks -alias &lt;your_key_alias&gt; to verify.</li>
<li>
Back Up Your Keystore: Immediately copy the .jks file to a secure location.</li>
</ul>
Exporting the Upload Certificate (Both Methods)
After generating the keystore, you need to export the certificate:
<ul><li>Command Line:</li></ul>
Bash
<pre>keytool -exportcert -keystore &lt;your_keystore_name&gt;.jks -alias &lt;your_key_alias&gt; -file upload_certificate.pem</pre>
This will create a `upload_certificate.pem` file.
<ul><li>
Android Studio: You can export the certificate from the same “Generate Signed Bundle/APK” dialog by clicking on it.</li></ul>
Requesting a New Upload Key in Google Play Console
<ul>
<li>
Go to App Integrity: In the Play Console, navigate to Release &gt; Setup &gt; App integrity.</li>
<li>
Go to App signing tab: Click on the App signing tab.</li>
<li>
Request Upload Key Reset: You’ll see an option to “Request upload key reset”. Click it.</li>
<li>
Follow the Instructions: Google will guide you through the process, which involves verifying your identity and ownership of the app.</li>
<li>
Upload the Certificate: Once approved, Google will prompt you to upload the upload_certificate.pem file you exported earlier.</li>
</ul>
Updating Your Flutter Project (If Applicable)
If you’re using Flutter, update your android/app/build.gradle file with the new keystore information:
Gradle
<pre>android { signingConfigs { release { storeFile file "&lt;path_to_your_new_keystore&gt;/&lt;your_new_keystore_name&gt;.jks" storePassword "&lt;your_new_keystore_password&gt;" keyAlias "&lt;your_new_key_alias&gt;" keyPassword "&lt;your_new_key_password&gt;" } } buildTypes { release { signingConfig signingConfigs.release } } }</pre>
Things to keep in mind.
<ul>
<li>
Security First: Keep your keystore file and passwords extremely secure. Losing them can cause significant problems.</li>
<li>
Backups are Crucial: Regularly back up your keystore file to a safe location.(your local desktop, a security key, a Version control system or any other place that’s easy for you to retrieve)</li>
</ul>
My hindsight is now your foresight..
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=379834a1dbbe" width="1" height="1" alt="">

How to Reset Your Upload Key for Google Play Store When It’s Lost, Compromised, or Forgotten

<h4>Navigating murky waters of banking in the US as an international resident.</h4>
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oexZ8jH4rVbIBOXhBTw2OA.png"><figcaption>An AI-generated image with prompt “proceed to generate an image for this title A founders guide to startup banking : navigating the murky waters of international banking in the US as a non-resident. Generate the image in a HI-RES format like Netflix movies; ultra realistic”</figcaption></figure>It is 4:44 PM EAT (East African Time) Saturday February 2025, I am seated in a hot room; rays of the sunlight stream through the window and the half-open door; the room feels stuffy and uncomfortable. A far-cry from the temperature regulated offices my friend runs that I normally go to in order to write code peacefully and in a comfortable environ. Today I decided to work from home because I didn’t suffer the morning chilliness gladly and all that traffic jam that any regular Gayaza-Kampala trekker/traveller has learnt to embrace and make peace with. Well I haven’t and never will.
My perilous albeit successful journey began 8 hours ago trying to find the perfect US banking partner for our startup <a href="https://intelliconcierge.com/">Intelli</a>. We are an accelerator/venture backed startup being supported by the wonderful portfolio team at MEST Africa. So we have the task of finding the best, most digital, startup-friendly, tech enabled and customer first bank of them all offering services like expense management, banking, corporate cards, treasury services, and venture debt (at the most sensible price).
A random search on google similar to this [US banks for startups] led me to this <a href="http://the%2017%20best%20startup%20banks%20and%20banking%20platforms%20for%20high-growth%20companies%20in%202025/">article </a>from Rho(which is surprisingly a fintech company, not a bank or an FDIC-insured depository institution according to the disclaimer notice on their footer.) I won’t list all the options in the article; you can read it for yourself.
To be clear we are fully incorporated in the state of Delaware although all our co-founders are non-US residents because of two reasons namely. 1. US incorporation makes the company access venture funding more easier because the company is registered in a country whose laws they are familiar with and thus makes oversight and due diligence super-easy for them 2. Who doesn’t want to list/go IPO on the NYSE when they hit unicorn status.
One of the other tremendous <a href="https://www.brex.com/spend-trends/financial-operations/mercury-alternatives-and-competitors">articles </a>I had to read as part of my journey to finding the best banking partner came from <a href="http://brex.com/">Brex </a>;
<h3>Choices, Choices, Choices</h3>
My first choice was Rho however it turned out they asked for some documents we didn’t as yet have. So I moved on to the next provider which was Brex; very beautiful platform; beautiful mobile app. Tremendous software. The onboarding process a breeze and delight. They asked for proof of funding(that would have to be our executed SAFE) it being a weekend where no one is in office I reckon I would have to go through a lot of high and low to get the document I need and the permission too in order to share this kind of information so I was stuck at that process. I had to move on and try another provider.
I tried <a href="https://www.grasshopper.bank/">grasshoper </a>which kind of reminded me of Mantis from the furious 5 in Kungfu Panda and sadly or rather to my dismay I saw an ugly notice “Sorry, you have been blocked You are unable to access mantl.com”
indicating to me that their services were not offered to people from my country (I wonder why). Anyway I moved on to the next best bank which was <a href="https://www.novo.co/">Novo</a>. The signup/sign-in process was very smooth i should say however I was quickly thrown off as when they required information for personal identification one of the fields required an SSN(Social Security Number) which I reckon is reserved for US citizens and residents. And Since I am not a US citizen. I picked up myself and moved on to the next service provider.
I tried all the major big boys; Chase Business Banking, First Republic Bank, JP Morgan, Bank of America and all that fall in this calibre and they require an SSN, a US number or US passport despite clearly stating that I am a non-us resident.
<blockquote>So I moved on to the next hoping with my fingers crossed that it might be my final stop. I signed up on <a href="https://relayfi.com/">Relay Financial</a>. When I saw their landing page I thought to myself “maybe these are cool guys” I mean there hero section has a bold statement Your cash flow can be clearer I finished the signup/sign-in smoothly and started the onboarding process. As I was close to wrapping up they requested for a US number… I was like okay that’s bye for me.</blockquote>
I tried out <a href="https://www.meow.com/">Meow </a>banking. The signup/sign-in was smooth we were in and ready to go. They went ahead too request us to link our US bank accounts but we didn’t have those yet because we were just starting out on our process of getting a bank account and that’s why we were in this position at this time. They didn’t require US numbers and addresses which I greatly appreciated. However I moved on because I was looking for something helpful for a startup trying to process and receive payments globally.
<h3>The Final Verdict</h3>
The final solution my team and I settled with was Mercury; oh sweet mercury.. They verified our EIN with our incorporation letter and the process was done; it’s now pending to be approved within 2–4 days and we will be good to go.
The journey to finding the right banking partner as a non-US resident founder has been an uphill one; It took my entire Saturday (or the better part of it) I couldn’t write code or do anything as I was doing correspondence with my CEO in Nairobi; switching between tabs and google searches while trying to find nuances hidden in fine print and all this work was just for us to open an account — it’s about time we get a player who is a digital bank that supports all nationalities to have access to US banking services and products indiscriminately and swiftly. We are in the 21st century.
All in all I am happy and please that we found a partner that understands the unique challenges of building a startup from outside the United States. After hours of research, applications, and countless “We’re sorry, but…” messages, I’ve learned some valuable lessons that I wish I’d known from the start and that’s why I wrote this article because someone out there might be in the same predicament I was in, may this be your lighthouse or buoy in the tempestous waters.
<h4>Why Mercury Emerged as the Winner</h4>
What sets Mercury apart isn’t just their willingness to work with non-US residents — it’s their entire approach to startup banking. Their verification process focuses on what matters: your company’s legal status and EIN, rather than the founders’ citizenship status. This approach acknowledges a fundamental truth about modern startups: innovation knows no borders.
The application process was refreshingly straightforward:
<ul>
<li>Submit your EIN and incorporation documents</li>
<li>Provide basic company information</li>
<li>Wait for 2–4 days for approval</li>
</ul>
No SSN required, no US phone number needed, and most importantly, no physical presence in the US necessary. For international founders like us at Intelli, this was a relief..
<h4>My hindsight and possibly your foresight</h4>
<ol>
<li>Start Early Don’t wait until you need to receive funding or make payments to start your banking search. The process can take time, and you want to have options rather than being forced to make rushed decisions.</li>
<li>Documentation is Key Keep your incorporation documents, EIN confirmation, and company formation papers well-organized and readily available. Having these at hand makes the verification process much smoother.</li>
<li>Research Banking Features for Growth While getting an account opened is the first hurdle, think about your future needs. Consider features like:</li>
</ol>
<ul>
<li>International wire transfer capabilities</li>
<li>API access for integration with your financial systems</li>
<li>Multi-user access for your team</li>
<li>Treasury management for when you raise funding</li>
</ul>
<h4>The Broader Perspective</h4>
This experience highlights a larger issue in the startup ecosystem: the growing disconnect between traditional banking requirements and the reality of modern, globally distributed startups. While the US remains a primary hub for startup incorporation and fundraising, the founders and teams building these companies are increasingly global.
<h4>Looking Forward</h4>
As we await our Mercury account approval, I can’t help but reflect on how this process could be improved for future founders. The financial industry is slowly adapting to the reality of remote-first, globally distributed teams, but there’s still much work to be done.
For those following in our footsteps, here are some key considerations for your banking partner search, may this article be a guide and light for you. And may things be as breezy for you as they were for me.
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=b5c61f7d2daa" width="1" height="1" alt="">

A founders guide to startup banking

<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Y-xKN2j0DMooWoHVHcnsVQ.jpeg"><figcaption>An AI-generated Image from Gemini of a corridor in a server room through multiple massive server racks.</figcaption></figure><h3>Introduction</h3>
I have scoured through hundreds if not thousands of google search results trying to find a free or atleast affordable option for a startup to host a backend on the internet so you don’t have to.
There are many hosting providers that are known to either actively support or work well with Django, including: <a href="https://www.heroku.com/">Heroku</a>, <a href="https://www.digitalocean.com/">Digital Ocean</a>, <a href="https://railway.app/">Railway</a>, <a href="https://www.pythonanywhere.com/">Python Anywhere</a>, <a href="https://aws.amazon.com/">Amazon Web Services</a>, <a href="https://azure.microsoft.com/en-us/">Azure</a>, <a href="https://cloud.google.com/">Google Cloud</a>, <a href="https://www.hetzner.com/">Hetzner</a>, and <a href="https://www.vultr.com/news/new-free-tier-plan/">Vultr Cloud Compute</a>. These vendors provide different types of environments (IaaS, PaaS), and different levels of computing and network resources at different prices.
But I assume you already know most of these. <a href="https://djangostars.com/blog/top-django-compatible-hosting-services/">Click Here</a> to read an article that does a comparison between most of the services/platforms/providers above. My mandate today is to indulge you in my personal handpicked providers that are intriguing to me.
<h3>My handpicked choices</h3>
<h4>1. Fly.io</h4>
<blockquote>“Scalable Full Stack Without the Cortisol”; is how they describe their value proposition on their landing page. Over 3 million apps have launched on Fly.io, boosted by global Anycast load-balancing, zero-configuration private networking, hardware isolation, and instant WireGuard VPN connections. Push-button deployments that scale to thousands of instances. To be fair i love or are impressed with them not only because of their snarky good copy but also the feel and UI/UX of their product.</blockquote>
The most important thing to discuss here is the pricing, so I have broken down a comprehensive summary of that; if you don’t feel like reading through it all; Here is your answer Current Hobby Plan: $5/month after free trial credit.
<h4>Fly.io Resource Pricing Summary</h4>
<h4>How It Works</h4>
<ul>
<li>
Billing: Per organization, based on provisioned resources, prorated by time.</li>
<li>
Organizations: Administrative entities for app development environments and billing management.</li>
</ul>
<h4>Plans</h4>
<ul>
<li>
Pay As You Go: Default plan for new organizations.</li>
<li>
Upgrade Options: Launch, Scale, Enterprise plans for more support and compliance options.</li>
<li>
Legacy Hobby Plan: No monthly fee, limited free resources.</li>
<li>
Current Hobby Plan: $5/month after free trial credit.</li>
</ul>
<h4>Free Allowances (Hobby, Launch, Scale Plans)</h4>
<ul>
<li>
Compute: Up to 3 shared-cpu-1x 256MB VMs.</li>
<li>
Storage: 3GB persistent volume.</li>
<li>
Data Transfer: 100 GB (North America &amp; Europe), 30 GB (Asia Pacific, Oceania, South America, Africa &amp; India).</li>
</ul>
<h4>Compute Pricing</h4>
<ul>
<li>
Running Machines: Charged based on CPU/RAM presets.</li>
<li>
Stopped Machines: Charged for root filesystem storage ($0.15/GB per month).</li>
<li>Example Prices:</li>
<li>shared-cpu-1x 256MB: $0.0000008/s ($1.94/mo).</li>
<li>performance-1x 2GB: $0.0000120/s ($31.00/mo).</li>
</ul>
<h4>GPU Pricing</h4>
<ul>
<li>On-demand GPUs:</li>
<li>A10: $1.50/hr.</li>
<li>L40S: $2.50/hr.</li>
<li>A100 40G PCIe: $2.50/hr.</li>
<li>A100 80G SXM: $3.50/hr.</li>
</ul>
<h4>Persistent Storage Volumes</h4>
<ul><li>
Cost: $0.15/GB per month.</li></ul>
<h4>Network Prices</h4>
<ul>
<li>
Dedicated IPv4: $2/month.</li>
<li>SSL Certificates:</li>
<li>Single hostname: $0.10/month.</li>
<li>Wildcard: $1/month.</li>
<li>Data Transfer:</li>
<li>North America &amp; Europe: $0.02/GB.</li>
<li>Asia Pacific, Oceania, South America: $0.04/GB.</li>
<li>Africa &amp; India: $0.12/GB.</li>
</ul>
<h4>Fly Kubernetes</h4>
<ul><li>
Managed Service: $75/month per cluster plus compute and storage costs.</li></ul>
<h4>Extensions and LiteFS Cloud</h4>
<ul>
<li>
Third-Party Services: Billed at list prices via Fly.io.</li>
<li>
LiteFS Cloud: $5/month for up to 10GB, $0.50/GB beyond.</li>
</ul>
<h4>Support</h4>
<ul>
<li>
Community Support: Included for all.</li>
<li>
Email Support: Available with Launch, Scale, Enterprise plans.</li>
</ul>
<h4>2. Koyeb</h4>
“Deploy to production, scale globally, in minutes. Accelerate AI apps with high-performance infrastructure. Connect your GitHub account to Koyeb. Choose a repository to deploy. Leave us the infrastructure — We build, deploy, run, and scale your application with zero-configuration.” Is how they describe their value proposition on their Landing page; I would say they are the dark horse in this race. Have tried deploying an app with them an its a breeze; connecting your Github repo and everything happens just smoothly;
The most important thing that you might be interested in not only their pricing but that fact that they have a <a href="https://tally.so/r/w4kzad">startup program </a>that you can apply to and get compute credit worth $30k. Additionally if you don’t feel like reading through the comprehensive pricing summary ; Here is your answer about Koyeb’s Pricing: Current Hobby Plan: $0/month pay-per-use No credit card required.
<h4>Koyeb Pricing Summary</h4>
<h4>Plans</h4>
Starter Plan
<ul>
<li>
Cost: $0/mo + compute (pay-per-use by the second).</li>
<li>Features:</li>
<li>1x free Web service.</li>
<li>1x free Postgres database (50h/month).</li>
<li>Run web apps, APIs, and workers.</li>
<li>Git push to deploy.</li>
<li>SSL and 10 free custom domains.</li>
<li>GPU access.</li>
<li>3 users included.</li>
<li>
Notes: No credit card required on Hobby plan; upgrade when ready.</li>
</ul>
Startup Plan
<ul>
<li>
Cost: $79/mo + compute.</li>
<li>Features:</li>
<li>Everything in Starter.</li>
<li>US, EU, and Asia regions.</li>
<li>6 users included (more as an add-on).</li>
<li>Up to 128GB of RAM &amp; 40 vCPU per service.</li>
<li>E-mail support and chat.</li>
<li>Slack cross-connect.</li>
<li>99.9% uptime SLA.</li>
</ul>
Enterprise Plan
<ul>
<li>
Cost: Custom, starting at $1000/mo with committed usage.</li>
<li>Features:</li>
<li>Custom RAM, CPU, and GPU.</li>
<li>Private dedicated locations.</li>
<li>Unlimited users.</li>
<li>SSO, RBAC, and Audit trail.</li>
<li>ISO27001 and SOC2 Certifications.</li>
<li>99.99% uptime SLA.</li>
<li>24×7×365 premium support.</li>
</ul>
<h4>General Features (Included in All Plans)</h4>
<ul>
<li>High-performance bare metal servers.</li>
<li>Global edge acceleration and CDN.</li>
<li>Autobuild for multiple languages (Node, Python, Go, Java, Ruby, PHP, etc.).</li>
<li>Global VPC across regions.</li>
<li>Free inbound bandwidth.</li>
<li>Deploy Docker from any registry.</li>
<li>HTTP/2, WebSocket, and gRPC support.</li>
<li>100GB of free outbound transfer.</li>
<li>Deploy from Dockerfile.</li>
<li>Service mesh &amp; discovery.</li>
<li>Anti-DDoS.</li>
<li>Infrastructure autohealing.</li>
<li>Automatic OS patching.</li>
<li>Continuous deployment with Canary.</li>
<li>Secrets management.</li>
<li>Horizontal scaling with global load-balancing.</li>
<li>CLI, API, Terraform, and Pulumi.</li>
<li>Real-time logs and metrics.</li>
<li>Autoscaling.</li>
<li>Community support.</li>
</ul>
<h4>GPU Pricing</h4>
<ul>
<li>
RTX-4000-SFF-ADA: $375/mo ($0.00014/s) — 20 GB VRAM, 6 CPU / 44 GB RAM.</li>
<li>
V100-SXM2: $632/mo — 16 GB VRAM, 8 CPU / 44 GB RAM.</li>
<li>
L4: $744/mo — 24 GB VRAM, 15 CPU / 44 GB RAM.</li>
<li>
L40S: $1488/mo — 48 GB VRAM, 30 vCPUs / 92 GB RAM.</li>
<li>
A100: $2009/mo — 80 GB VRAM, 15 vCPUs / 180 GB RAM.</li>
<li>
H100: $2454/mo — 80 GB VRAM, 15 vCPUs / 180 GB RAM.</li>
</ul>
<h4>Serverless Postgres Pricing</h4>
<ul>
<li>
Free: 50h/month, 1 GB RAM, 0.25 CPU.</li>
<li>
Small: $21/mo, 1 GB RAM, 0.25 CPU.</li>
<li>
Medium: $42/mo, 2 GB RAM, 0.5 CPU.</li>
<li>
Large: $95/mo, 4 GB RAM, 1 CPU.</li>
<li>
Additional Storage: $2/mo per 10 GiB.</li>
<li>
Usage: $0.000008/s for Small; $0.10/GB for additional storage.</li>
</ul>
<h4>FAQs</h4>
<ul>
<li>
Credit Card: Not required to start on the Hobby plan.</li>
<li>
Starter Plan: Free, with one Web Service and Database, 10 custom domains; credit card required.</li>
<li>
Charges: Based on actual usage, billed per second.</li>
<li>
Differences Between Plans: Starter for non-critical apps; Startup for production; Enterprise for large, mission-critical apps.</li>
<li>
Outbound Bandwidth: Charged at $0.04/GB after free tier.</li>
<li>
Custom Domain: Yes, with free SSL for 10 domains.</li>
<li>
Cancellation: Anytime with prorated credit for unused time.</li>
</ul>
<h4>Render</h4>
Finally darling Render 😍😘; This has been the go-to platform for most of our MVP projects; we’ve tried it; tested it and stretched the limits of it’s free resources and are forever grateful for the team at Render. It’s no mean feat giving free compute albeit small or miniscule to users indefinitely. (Not speaking about the database).
Render describes their value proposition as “Your fastest path to production. Build, deploy, and scale your apps with unparalleled ease — from your first user to your billionth.” And yes you can get started for free.
Here is a comprehensive breakdown of their pricing if you want to go over the free fence and get into the real thick of things.
<h4>Render Pricing Summary</h4>
<h4>Plan Tiers</h4>
Individual
<ul>
<li>
Cost: $0/month + compute costs.</li>
<li>
Designed for: Hobbyists, students, and indie hackers.</li>
</ul>
Team
<ul>
<li>
Cost: $19/user/month + compute costs.</li>
<li>
Designed for: Small teams and early-stage startups.</li>
</ul>
Organization
<ul>
<li>
Cost: $29/user/month + compute costs.</li>
<li>
Designed for: Larger teams with complex needs.</li>
</ul>
Enterprise
<ul>
<li>
Cost: Custom pricing + compute costs.</li>
<li>
Designed for: Ultimate power and customization needs.</li>
</ul>
<h4>Services</h4>
<ul>
<li>Static Sites</li>
<li>
Cost: $0/month.</li>
<li>
Features: Lightning-fast CDN, automatic continuous deploys from Git, instant cache invalidation, custom domains with fully managed TLS.</li>
<li>Web Services</li>
<li>
Cost: From $0/month.</li>
<li>
Features: Web services with HTTP/2 and full TLS, support for Node, Python, Go, Rust, Ruby, Elixir, private services, custom Docker containers, SSDs at $0.25/GB per month.</li>
<li>PostgreSQL</li>
<li>
Cost: From $0/month.</li>
<li>
Features: Fully managed PostgreSQL, connect from anywhere, automatic daily backups for paid instances, expandable storage, 1 GB SSD storage included, point-in-time recovery.</li>
<li>Redis®</li>
<li>
Cost: From $0/month.</li>
<li>
Features: Fully managed Redis, connect from anywhere, maximize availability by queueing jobs, reduce primary DB’s load, cache pages, results, fragments.</li>
<li>Cron Jobs</li>
<li>
Cost: From $1/month.</li>
<li>
Features: Run any command or script periodically, failure notifications, full support for cron expressions, prorated by the second, live cron job logs.</li>
</ul>
<h4>Features Across Plans</h4>
<ul>
<li>
Build &amp; Deploy: Web Services, Static Sites, Cron Jobs, One-Off Jobs, Background Workers, Free Instances, Build Pipeline (500 minutes/month for Individual, 500 minutes/user/month for Team and Organization, custom for Enterprise).</li>
<li>
Free Bandwidth: 100 GB for Individual, 500 GB for Team, 1 TB for Organization, custom for Enterprise.</li>
<li>
Data Stores: Managed PostgreSQL, Automated Backups, Read Replicas, Point-in-Time Recovery, High Availability, Managed Redis, Persistence, Persistent Disks.</li>
<li>
Collaboration: Auto Deploys from Git, Service Previews, Deploy Hooks, Preview Environments, up to 10 team members for Individual, unlimited for Team and Organization, custom for Enterprise.</li>
<li>
Convenience: Native Runtimes, Docker Builds, Pre-Deploy Command, Infrastructure-as-Code, Environment Groups, Monorepo Support, Automatic Brotli Compression, Custom Domains with free TLS, Wildcard Domains.</li>
<li>
Uptime &amp; Monitoring: Zero-Downtime Deploys, Health Checks, Instant Rollbacks, Log Retention (7 days for Individual, 14 days for Team, 30 days for Organization and Enterprise), Log Streams, HTTP Request Logs, Contractual Uptime SLA.</li>
<li>
Networking: Premium Global CDN, HTTP/2 and HTTP/3, WebSocket, Static Outbound IPs, Private Networking, Service Discovery.</li>
<li>
Security: HTTPS only, SSH access, DDoS Protection, Secret Files, 2FA, GDPR DPA, SOC 2 Type 2 Report.</li>
<li>
Support: Community Forum, Email Support, Chat Support, Migration Support, Customer Success Engineer, First Response Support SLA, Urgent Support SLA, Architecture Review, Dedicated Support Engineer, Private Slack Channel, Quarterly Account Review.</li>
</ul>
<h4>Billing and FAQs</h4>
<ul>
<li>
Billing: Monthly plan costs depend on plan type and number of users, compute usage is billed and prorated by the second.</li>
<li>
No Activity Fee: Team member fee waived if no services and no activity during a month.</li>
<li>
Team Members: Billed based on user count at the end of the billing period.</li>
<li>
Pipeline Minutes: Track build and pre-deploy task durations, viewable from the dashboard’s Billing page.</li>
<li>
Free Instances: Usage limits designed for exploring tech, building personal projects, and previewing Render’s experience.</li>
<li>
Payment Methods: Accepts all major credit and debit cards, processed securely by Stripe.</li>
</ul>
<h3>Some of the things to consider when choosing a host:</h3>
<ul>
<li>How busy your site is likely to be and the cost of data and computing resources required to meet that demand.</li>
<li>Level of support for scaling horizontally (adding more machines) and vertically (upgrading to more powerful machines) and the costs of doing so.</li>
<li>Where the supplier has data centres, and hence where access is likely to be fastest.</li>
<li>The host’s historical uptime and downtime performance.</li>
<li>Tools provided for managing the site — are they easy to use and are they secure (e.g. SFTP vs. FTP).</li>
<li>Inbuilt frameworks for monitoring your server.</li>
<li>Known limitations. Some hosts will deliberately block certain services (e.g. email). Others offer only a certain number of hours of “live time” in some price tiers, or only offer a small amount of storage.</li>
<li>Additional benefits. Some providers will offer free domain names and support for TLS certificates that you would otherwise have to pay for.</li>
<li>Whether the “free” tier you’re relying on expires over time, and whether the cost of migrating to a more expensive tier means you would have been better off using some other service in the first place!</li>
</ul>
<h3>Conclusion</h3>
That’s the end of this article on where to host your Django (or any) backend and also my cherry picks for providers. I hope you’ve found them useful. You can check out an open source Project Management software we are building on <a href="https://github.com/orgs/Clearpath-HQ/repositories">GitHub here</a>. Contributions and any sort of support is welcome.
<h3>See Also</h3>
<a href="https://docs.aws.amazon.com/elasticbeanstalk/latest/dg/create-deploy-python-django.html">AWS Beanstalk tutorial</a>
<a href="https://fly.io/django-beats/deploying-django-to-production/">Deploy Django on Fly.io</a>
<a href="https://cloud.google.com/python/django/appengine">Deploy django app on Google App Engine</a>
<a href="https://realpython.com/django-hosting-on-heroku/">Hosting a Django project on Heroku</a>
<a href="https://docs.render.com/deploy-django">Deploy a Django app on Render</a>
<a href="https://www.geeksforgeeks.org/how-to-deploy-django-project-on-pythonanywhere/">Deploy a Django project on pythonanywhere</a>
<a href="https://developer.mozilla.org/en-US/docs/Learn/Server-side/Django/Deployment">Documentation on Django deployment</a>
<a href="https://platform.sh/blog/how-to-deploy-your-django-project-on-platformsh/">Deploying django backend on platform.sh</a>
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=3f4f9f594aa5" width="1" height="1" alt="">

Where to host your Django (or any) backend.

<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_RvseKGMLyGFnZFWOYT-aA.jpeg"><figcaption>An AI generated image from Gemini using prompt : “ a dark book cover with a processor chip in the background and electronic circuitry, with title Getting Started with Architecture for AI apps: A founder’s guide by Were Samson Bruno”</figcaption></figure>This is a personal reflection of my journey, tools I have come to become familiar with and gradually refined as I kept on venturing deeper into the rabbit hole of AI architecture
<h3>Introduction</h3>
I assume that you already possess a keen interest in Artificial intelligence and you are now excited and want to build your own AI app. It could be a ChatGPT wrapper or something entirely unique that has never been applied in any industry before. This journey you desire to walk I have walked and will take you where I have been.
For reference purposes I will use the initial architecture of our solution <a href="https://www.intelliconcierge.com/">Intelli</a>. Our team is made up of four individuals — each with a prefix appended to the Role CTO, which means everyone’s opinion influences the technical decisions we make. When we envisioned Intelli(fondly referred to as Elli) we saw it this way; We broke it down into movable parts that would come together to form the sweet spot of Data and Intelligence that would solve our customer’s problems in the easiest and most cost-effective way
<h3>OSI Model Diagram</h3>
<figure><img alt="The OSI Model (from Top to Bottom) of Intelli Tech Architecture" src="https://cdn-images-1.medium.com/max/854/1*TgAqJbfkwgqNoK003_IGwg.png"></figure>Bare in mind that the decisions for the various platform choices were made based off of cost, performance, ease of use, our team’s acquaintance with them and community built around it, for example we had Azure credits worth $1,000 and we had google credits worth $200; It only made financial sense that we rely on Azure although we acknowledge it was not the most friendly cloud platform available.
For the choice on <a href="https://www.oracle.com/database/what-is-database/">databases</a>, we segmented them into two usecases; one for basic storage and retrieval; We went for <a href="https://www.cloudflare.com/learning/ai/what-is-vector-database/">a vector database</a> because they allow for fast similarity searches; which is what we wanted since we had a RAG model as part of our architecture at a point. We knew there was going to be indexing of data and a vector database would be the best way to store it.
Yet again the decision for what we chose came down to available resources and familiarity with the tech; we found <a href="https://docs.pinecone.io/reference/create_index">documentation</a> on how to setup indices in Pinecone and also saw some <a href="https://python.langchain.com/docs/integrations/vectorstores/pinecone">documentation on Langchain</a> specifically tailored for pinecone. Adding to the fact that we ran into some errors trying to use Redis which was our first choice for a vector DB; we quickly switched our minds to Pinecone.
MongoDB and <a href="https://www.matillion.com/blog/what-is-azure-data-lake">Azure Data Lake</a> didn’t get the light of day for different reasons; The former didn’t have the easiest path to entry despite having plenty documentation out there; It had a steep learning curve and thus we chose to forego it. However much we acknowledge the robustness of the <a href="https://www.mongodb.com/developer/languages/python/semantic-search-made-easy-langchain-mongodb/">MongoDB Atlas Semantic search</a> and <a href="https://www.mongodb.com/developer/products/atlas/rag-atlas-vector-search-langchain-openai/">coupling </a>it with a <a href="https://aws.amazon.com/what-is/retrieval-augmented-generation/">RAG model</a>.
The latter was too early for us. We aren’t yet at the level, or scale of having big data and as such would just be a wastage of resources for us to run this service if we have failed to exhaust our current DBs. Although we shall still keep it on our rack for when we get there and our needs warrant such storage capacities.
This brings me to the second segment of our Dbs; Data Mining and Machine Learning; we saw the essence of having our system become better at answering questions by fine-tuning a foundational model to turn it into a proprietary and enterprise LLM specifically built for our business usecase and fully decoupled from and independent of it’s source/parent model.
When we started we only thought about using PyTorch as the library for Natural Language Processing(NLP) within <a href="https://colab.google/">Google Cola</a>b as the playground for us to perform our training but our minds were introduced to better options like TensorFlow which is an End to end open source platform for machine learning and thus would serve our needs adequately and ticks most of the boxes we needed.
Power BI (in Microsoft Fabric) were basically reporting tools; to visually represent the data/numbers of the progress of our training and Machine Learning procedures.
Moving up to the Presentation Layer; we chose to stick to A Python Back-end using the Django framework for it’s benefits of being structured, organized and an out of the box solution(comes with batteries included) It greatly increases our speed of building. And the fact that we also had access to technical expertise in our vicinity that were proficient with it made it a sound decision to make. The most important reason for this decision however is because we thought about the intricacies of dealing with data for ML purposes in any language other than python and it made sense that we work with python from day 0.
We chose to host our codebase on GitHub and initially had decided to ship it to production in a docker container; however as we proceeded we found out alternative ways to ship it that would be easier for us to execute. Vercel provided a serverless edge option to launch our Django app and we lunged at it(I love all things serverless!).
For the APIs we are utilizing Retell for Voice, Paddle for Payments, Our LLMs are OpenAi, Llama 2 and hope to soon have our own LLM Epsilon join these ranks.
For the Presentation Layer we chose to use NextJs framework to build the web app and Flutter to build the desktop for reason of our team having a familiarity and experience with these tools and technologies and thus making it a perfect fit.
<h3>Technical Architecture Diagram</h3>
Now that I have explained the OSI model; you can glimpse at the architecture and it will make sense to you. As you look at the architecture diagram you will realise there are elements that were added that were not part of the OSI model; this is because we kept on refining and changing our architecture as we learned and kept on going.
<figure><img alt="A diagram of our technical architecture from frontend(left) to backend(right)" src="https://cdn-images-1.medium.com/max/887/1*0rOiuC6qBPhX135ngucbFw.png"></figure><h4>Additional Notes</h4>
Some of the components we switched entirely are listed herein:
<ol>
<li>Cloud Service Provider — Much as we love GCP; and had crafted our entire cloud infrastructure around them. We had to choose to flow where the credits flowed, to make our building azurier(easier)</li>
<li>Claude and Palm2 — We thought we would start with these two models but things turned out Open AI, so we took that path.</li>
<li>We had missed a crucial part (<a href="https://www.altexsoft.com/glossary/online-travel-agency-ota/">OTAs</a>) so we had to look out for those that had developer platforms and provided their API services.</li>
</ol>
<h3>Conclusion</h3>
I believe if you have followed this article to this point, your mind has been opened up to see things as they are; plain. I hope this has been an insightful read for you and that it has inspired you to come up with an even more optimised architecture and subsequently product.
This was written to help you see AI app architecture from the angle of any other software you have built before; what changes are the tools and the application of the software and the paradigm of building it.
You can build anything you set your mind to, so go forth and build.
Edit : I am changing the cover of the article to just show the improvement AI has gone through. kind reminder.
<blockquote>“It can only get better. This is the worst it will ever be.”</blockquote>
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=a26bf8f16dad" width="1" height="1" alt="">

Getting started with Architecture for AI apps: A Founder’s guide

<h3>ChopCompass: How I built an AI-Powered Chatbot in 30 minutes to Discover the Best Dining Experiences</h3>
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VPpnpm_czLy68zDYonQVbw.png"><figcaption>Chop Compass; a glimpse</figcaption></figure><h3>Introduction</h3>
On the bus to MEST campus today(2nd bus). I sat next to a colleague <a href="https://medium.com/u/7ef90d577d96">Andrews Kwesi Ankomahene</a> and we had a light chat about a recently launched hackathon by AWS on devpost. The PartyRock Generative AI Hackathon by AWS: a challenge for all builders to get hands-on making generative AI-powered apps. In this challenge, we are required to use PartyRock, an Amazon Bedrock Playground as a fast and fun way to learn about Prompt Engineering and Foundational Models (FMs) to build a functional app with generative AI.
We are also expected to experiment with prompts to build an app based on one of four challenge categories or Remix the featured Side Quest app for more ways to compete!
The challenge categories are below:
<ul>
<li>Interactive Learning Experience — virtual science lab, language lesson, etc;</li>
<li>Creative Assistant — music composition assistance, story idea generator, etc;</li>
<li>Experimental Entertainment — interactive story, Role Playing Games, etc;</li>
<li>Freestyle — any project that doesn’t fall into one of the categories above.</li>
</ul>
You can either create a Party Rock app from scratch or Remix an existing one. I chose to do the latter being that it was already looking like what I envisioned.
<h3>Backstory and genesis!</h3>
Discovering the perfect dining experience can be a challenge but not anymore. As a firm believer in innovation and reinvention, I was inspired by the convenience and personalized recommendations offered by Google Maps. This inspiration led me on a journey to develop an AI-powered chatbot that would revolutionize the way people find the best restaurants.
Welcome to the story of ChopCompass, my simple chat bot solution that quickly helps you in discovering exceptional dining experiences based on your location, budget, cuisine preference, and meal you want to eat.
<h3>Finding Inspiration in Google Maps</h3>
The idea for ChopCompass was born out of the desire to create a user-friendly platform that could provide personalized recommendations for dining establishments. Seeing how Google Maps effortlessly provided directions and suggestions based on user preferences, I recognized the immense potential for a similar concept in the food industry. With this inspiration in mind, I wanted to develop an AI-powered chatbot that would take restaurant discovery to the next level.
<h3>Developing ChopCompass: A Simple Walk through</h3>
<ol><li>I visited The <a href="https://partyrock.aws/">Party Rock website</a> as shown in the image below</li></ol>
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HbEIM11mu_aHEIobnhXs8Q.png"></figure>2. I scrolled to the section (Try out featured apps built with PartyRock)
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KAC8i9yu3Ov1putVOfUp9Q.png"></figure>3. I clicked on the Good Eats App because I had something similar in mind and when I did; I was redirected to the page displayed below: This is because I already had created a Party Rock account. But if you don’t have an account it will prompt you to create one and you can use apple or Google SAML.
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bDgg8yAkvAYDNpXkCJoumw.png"><figcaption>Good Eats Restaurant App Landing Page</figcaption></figure>4. I clicked on the Remix button In the top Right corner
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uu8o46HiNxZa0x17frPT2w.png"></figure>Immediately I did; it spun up my own instance of the Good Eats App.
5. From here on i added widgets; the first widget I added was the budget widget
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hGXOntWm64h-rAGL6iUSxg.png"></figure>6. After adding the widget; I proceeded to save the changes and published the app; subsequently you have to keep on pressing the release changes button after clicking save
7. One important feature I loved was the Show Configuration button on each of the widgets; It helps you choose the LLM you want to use and set the parameters like temperature and Top.
<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I3CppRFkMNGElpSXwkv54g.png"></figure>8. Finally once the app was ready I launched it.
<h3>What Next:</h3>
You can find Chop Compass <a href="https://partyrock.aws/u/Samsonroyal/C4jLd2-Tb/Chop-Compass">here </a>and test it out
Side Note: While developing ChopCompass, I love the fact that AWS party rock comes with about 7 open-source base LLMs and the UI of the site is leaning towards Neo-brutalism. I love it.
<h3>Testing and Iteration: Refining ChopCompass</h3>
Once the development was complete, it was time to put ChopCompass to the test.
User feedback plays a vital role in refining the functionality of products and ensuring their effectiveness. I shared a link to the remix app in our group of fellow Global entrepreneurs so they could get to test it out and use it extensively.
I got some exciting feedback and also got two people to remix my app and go ahead to create their own apps. The feedback they gave also enabled me to identify areas for improvement and address any usability issues.
<h3>Reflection and Future Enhancements</h3>
Creating ChopCompass was a swift and exciting process that took only 30 minutes. I utilized the no-code platform, AWS Party Rock, to build the app. I started by remixing an existing app, adding more widgets, and making a few edits to customize it to my needs. Within a short time, I had successfully spun up ChopCompass on the AWS Party Rock platform.
The development process was straightforward, and I encountered no significant challenges along the way. ChopCompass was built with a focus on user-centric design, leveraging the convenience of AI-powered chatbots and the personalized recommendations of Google Maps. The app’s core features include the ability to provide tailored dining suggestions based on user preferences such as location, budget, cuisine type, and meal type.
My experience building ChopCompass has been both educational and rewarding. I learned the importance of prioritizing user experience and the potential of AI in transforming everyday experiences. I was able to efficiently create a unique and functional app in a short amount of time using a no-code platform.
Moving forward, I plan to enhance ChopCompass by incorporating advanced machine learning algorithms to improve the chatbot’s recommendations and expand its reach.
I believe that ChopCompass has the potential to revolutionize the way people discover dining experiences and look forward to sharing my journey with you.
TL:DR
<ul>
<li>ChopCompass was developed in a <a href="https://awspartyrockhackathon.devpost.com/">hackathon</a>.</li>
<li>ChopCompass was built entirely on a no-code platform called AWS party rock</li>
<li>It utilises the Claude LLM</li>
<li>The chatbot’s recommendations are based on a combination of user cuisine preferences, location, and budget.</li>
<li>The app allows users to scout for and find good diners and restaurants they can visit and patronise with friends.</li>
</ul>
<img src="https://medium.com/_/stat?event=post.clientViewed&amp;referrerSource=full_rss&amp;postId=451c3e0df60d" width="1" height="1" alt="">

Informative

Recent Articles