DEV Community

Cover image for Gemini Became My Entire Hackathon Team — How a Solo Dev in Kenya Won His First MLH Prize Building RepoX
Cian
Cian Subscriber

Posted on

Gemini Became My Entire Hackathon Team — How a Solo Dev in Kenya Won His First MLH Prize Building RepoX

Built with Google Gemini: Writing Challenge

This is a submission for the Built with Google Gemini: Writing Challenge

What I Built with Google Gemini

Picture this: It’s 3 a.m.. My first-ever MLH hackathon. I’m staring at a blank screen, heart racing, knowing I’m completely outmatched by teams with years of experience.

Then I opened Google Antigravity — and everything changed.

I built RepoX: an interactive platform that turns any public GitHub repository into a living, breathing learning adventure.

No more getting lost in massive codebases. RepoX gives you:

  • A stunning D3.js force-directed graph that maps every file and its relationships like a neural network
  • Instant AI-powered explanations for any file (including “Explain Like I’m 5” mode that actually makes sense)
  • Smart personalized learning paths — the AI reads the entire repo and tells you the exact smartest order to explore it
  • Progress checklists and history so you never lose momentum

The crazy part? The app itself runs on Gemini. Every explanation and learning path is generated live by the Gemini API (securely routed through Cloudflare Workers).

And yes — this project won Best AI Application Built with Cloudflare at Hacks for Hackers 2026. My very first hackathon… and I took home a prize. I still can’t believe it.

Demo

Live app (paste any GitHub repo and watch the magic): https://main.repox.pages.dev

Full YouTube demo:

Devpost: https://devpost.com/software/repox

GitHub: https://github.com/mutaician/RepoX

What I Learned

This wasn’t just a hackathon project — it was my crash course in what happens when you stop coding alone and start coding with an AI teammate.

I went from zero D3.js experience to building a smooth, responsive graph that handles thousands of nodes. I learned secure API proxying on Cloudflare Workers under extreme time pressure. I mastered prompt engineering at a level I never thought possible — crafting system prompts so precise that Gemini would output perfectly formatted learning paths every single time.

Most importantly, I learned that one determined developer + the right Gemini workflow can outpace entire traditional teams. The confidence this gave me is something no tutorial could ever provide.

Google Gemini Feedback

Gemini wasn’t a tool. It was my co-founder, my senior dev, my QA tester, and my creative director — all in one.

I used Antigravity (Google’s agentic IDE) the entire time:

  • Gemini 3 Pro handled the heavy lifting — autonomously designing the learning-path algorithm, reasoning through complex repo analysis, and even suggesting UI tweaks that made the graph feel alive.
  • Gemini 3 Flash was my speed demon — instantly generating UI components, ELI5 explanations, and quick fixes while I kept momentum.
  • Gemini 2.5 was the reliable fallback when context got too big on massive repos.

What blew me away:
The agentic flow was unreal. I’d describe a feature once, and Antigravity would plan, code, debug, and iterate — often better than I would have done myself. The personalized learning paths Gemini 3.1 created were scarily good — logical, educational, and genuinely helpful.

Where it got messy (keeping it real):
Larger repos sometimes overwhelmed the context window and the agent would start hallucinating relationships or going off on wild creative tangents. I had to get surgical with my prompts and occasionally switch models. Response formatting could be inconsistent (markdown breaking in weird places), and yes, the token costs added up during heavy 3.1 sessions.

But here’s the truth: Without this exact multi-model + Antigravity setup, RepoX would still be a half-finished idea on my laptop. Gemini didn’t just help me finish — it helped me win my first hackathon.

From a nervous solo dev in Kenya to MLH prize winner in 48 hours. That’s the power of Google Gemini.

Thanks for reading my story — can’t wait to see what we build next. 🚀

Top comments (2)

Collapse
 
nadinev profile image
Nadine

Interesting project!

Collapse
 
kaiav_nihalani profile image
Kaiav Nihalani

Great sounding project! I'm just wondering, as a fellow google antigravity user, did you use any antigravity plugins like Ralph Loop or GSD for antigravity? I find that GSD for antigravity especially improves the experience working with antigravity