What is RepoDoc AI?
Every developer skips documentation. Writing a README, Changelog, and Contributing Guide takes hours — so most projects ship without them. RepoDoc AI fixes that in seconds.
Paste any public GitHub repository URL. The app reads your commits, file tree, and metadata — then writes everything for you. No login. No setup. No CLI.
What It Generates
- README.md — Professional, badge-included, with features, installation, usage, and tech stack
- Changelog — Human-readable, grouped by features, fixes, and chores from real commit history
- Contributing Guide — Fork instructions, branch naming, commit style, PR process
How I Built It
Built as a single HTML file — no React, no Node.js, no build step.
- GitHub REST API fetches repo metadata, file tree, config files (package.json, CMakeLists.txt, Cargo.toml), and last 10 commits
- Groq API (llama-3.3-70b-versatile) generates all three documents in parallel
- Auto-fallback system — if the selected model hits rate limits, the app silently retries with the next model
- Markdown Preview + Raw toggle on every tab
- Download All as .md in one click
Challenges
Building a context pipeline that gives the AI enough information without overloading the context window required intelligent filtering — prioritizing structure and commits over raw source code.
Handling rate limits gracefully — without showing raw API errors — required abstracting retry logic entirely behind the UI.
What I Learned
- Fault-tolerant API pipeline architecture in Vanilla JS
- Structured prompt engineering for consistent Markdown output
- Building a premium UI without any CSS framework
What's Next
- GitHub App integration — auto-open PRs when docs go stale
- Private repo support via OAuth
- VS Code extension
- Export to Docusaurus / MkDocs ```
Log in or sign up for Devpost to join the conversation.