My name is Sumama Jameel. I am 14 years old and I live in Gulshan-e-Zia, Orangi Town, Sector 11 1/2, Karachi, Pakistan. I study at Tameer-e-millat model school, Karachi. You can reach me at sumamajamil2005@gmail.com and you can find everything I have built on GitHub at github.com/Sumama-Jameel.
I want to tell you who I am, what I do, and how I work. I am not going to dress it up or make it sound more impressive than it is. I am going to give you the full picture so you can decide for yourself whether my work is worth your attention.
What I Do
I build software products from the ground up. When I say I build them, I mean I handle the entire lifecycle of a product. I research the problem space to understand what is broken and who is suffering from it. I architect the system by deciding how every component should interact, what technologies should be used, and what constraints must be enforced. I plan the implementation by breaking the architecture into small, testable pieces that can be built one at a time. I direct the engineering process, review every output, reject what does not meet my specifications, and iterate until the product works exactly the way I designed it to work.
I do not outsource the thinking. The architecture is mine. The constraints are mine. The verification criteria are mine. The product exists because I understood the problem deeply enough to design a solution that actually solves it.
I have shipped nine products this way. Some of them are large systems with multiple components. Some of them are focused tools that do one thing well. All of them are open source and all of them solve a problem I personally encountered.
The Products I Have Built
Megabit is a data generation engine for AI coding models. The problem it solves is that AI labs need training data that shows the full process of software development, not just the final code. A git diff shows you what changed but it does not show you how the developer explored the codebase, made a plan, wrote code, ran tests, hit errors, debugged them, and eventually got it right. Megabit reconstructs that process. It takes a real code change from an open-source repository, sets up a sandbox with the before state of the code and the after state of the tests, hides the actual solution, and makes an AI agent recreate the change from scratch. Every command the agent runs is actually executed in a Docker container. Every error is real. Every test result is real. The system then verifies the agent's output against the real human solution and only keeps episodes that pass verification. I researched the synthetic data market, I architected the anti-faking sandbox, I planned the verification pipeline, and I built the entire system.
Mirage is an AI gateway that gives you a single OpenAI-compatible endpoint routing to ChatGPT, DeepSeek, Gemini, Qwen, Mistral, and Meta AI without requiring API keys or per-token billing. It works by importing your existing browser sessions from Chrome, Firefox, or Edge and using the same internal web APIs that power the chat interfaces you already use. I built it in Rust using a custom headless browser engine called Obscura. The system maintains a pool of ten warmed browser contexts, handles rate limiting and circuit breakers, and supports streaming, tool calling, web search, and file uploads. I researched how the internal APIs work, I architected the session management system, I planned the multi-provider routing, and I built the entire gateway.
Grasper is a reading companion for hard science books. It solves the problem where people read a page of a physics or engineering textbook, think they understand it, move on, and forget most of it within a day. Grasper shows you one page at a time, generates a first-principles explanation of that page using an AI model you choose, asks you five comprehension questions, and only unlocks the next page when you answer four out of five correctly on the first try. It supports PDF, EPUB, and plain text files. It renders equations properly with KaTeX and has a button that explains each symbol in an equation. It can decode diagrams and figures using a vision model. Each page can have its own chat thread for follow-up questions. Your book stays on your machine and only the single page you are studying gets sent to the AI. I researched learning science and the forgetting curve, I architected the comprehension gate system, I planned the multi-format parsing pipeline, and I built the entire application.
Validator is an agent skill that validates a product or project idea before you build it. It researches the market, finds every competitor including enterprise tools, startups, open source projects, and side projects, destroys your assumptions, checks whether your supposed unique feature already exists, evaluates your moat, checks platform policy and technical feasibility, and returns a GO, CONDITIONAL GO, or NO GO verdict backed by sourced evidence. Every claim in the report carries a source URL. If the verdict is CONDITIONAL GO, it states a specific falsification test that would flip the verdict to NO GO. It forces you to pick one target audience and rejects vague answers like "all businesses" or "everyone." I researched the validation methodology, I architected the twelve-phase research process, I planned the evidence hierarchy, and I built the entire skill by myself. Manually.
Whelmer is a browser debugging tool that gives AI agents full visibility into what a browser is doing. It connects to Chrome, Brave, Edge, Opera, and Firefox using CDP, BiDi, FDP, and Foxbridge protocols. It shows every DOM change, every network request, every console message, every storage mutation, and every security state change. Nothing is hidden and nothing is filtered. It keeps a ten-thousand-event ring buffer so you can replay history. It supports recording sessions, replaying captures as HAR or CSV or JSON, diffing two captures to find what changed, and rebuilding recorded requests as curl commands. It is written in pure Rust as a single binary with no Node.js, no Python, and no npm install required. I researched the browser debug protocols, I architected the event bus and multi-protocol connection system, I planned the capture and replay engine, and I built the entire tool.
SubZero is a privacy-first subscription tracker that finds hidden recurring charges in bank statements without connecting to your bank. It parses statement files locally and identifies subscriptions, trials, and recurring payments. I researched the subscription detection patterns, I architected the parsing system, and I built the tool.
How I Work
My process for every product follows the same structure. I start by researching the problem until I understand it completely. I read existing solutions, I find where they fail, I identify what is missing, and I define exactly what needs to exist for the problem to be solved. Then I architect the system. I decide what components are needed, how they talk to each other, what the data flow looks like, what the constraints are, and what the failure modes are. Then I plan the implementation by breaking the architecture into ordered steps where each step produces something testable. Then I build it, piece by piece, testing each piece before moving to the next.
I use AI as my engineering team. I write the specifications, I define the interfaces, I set the constraints, and I direct the implementation. I review every line of output. I reject what is wrong. I ask for fixes. I test the result against my specification. This is not a shortcut. This is a method. The thinking, the architecture, the planning, and the verification are all mine. The typing is delegated.
What I Am Working On Right Now
My primary focus is Megabit. I am generating sample datasets to send to AI labs, I am applying to Y Combinator for the Winter 2027 batch, and I am building the sales pipeline with my business co-founder. The goal is to become a supplier of verified coding trajectory data for AI companies. The AI training data market is over four billion dollars in 2026 and it is growing at over twenty percent per year. AI labs need grounded data that shows real development processes. That is what Megabit produces.
Connect With Me
If you are working on AI training data, if you are building coding assistants, if you are interested in any of my open source projects, or if you want to discuss system architecture and product design, reach out.
Email: sumamajamil2005@gmail.com
GitHub: github.com/Sumama-Jameel
Location: Karachi, Pakistan
I respond to every message.
Top comments (0)