ungithub

⭐️ AI-powered codebase explorer. Ungithub helps you deeply understand any open-source project with the help of AI. 🚀

Comment

🧠 ungithub

AI-powered codebase insight explorer - <!-- just replace github.com with ungithub.com (DOMAIN CURRENTLY NOT PURCHASED) --> Go to ungithub.vercel.app, to unlock instant, smart analysis of any public repository.

Example:
🔗 https://github.com/jayanth920/t2s-s2t
becomes
🔍 https://ungithub.vercel.app/jayanth920/t2s-s2t

  • Indexing takes just a few seconds to minutes depending on the repository size and backend startup. Once indexed, answers are delivered instantly.

🚀 What is ungithub?

ungithub helps you deeply understand any open-source project with the help of AI. Whether you're onboarding to a new codebase or just exploring, simply paste a GitHub URL <!-- (or replace github with ungithub) (CURRENTLY DOMAIN NOT PURCHASED, GO TO THE Website) -->to paste repo url and get:

  • 🔍 Auto-summarized repo insights
  • 🧠 AI-generated explanations of files, folders, and logic
  • 🧩 Embedding-based search across the entire repo

- 📎 Ask questions about the code like ChatGPT - but repo-specific

Youtube Video

Youtube Video

🛠️ Tech Stack

Layer Tech
Backend Python, FastAPI
AI & Embedding OpenAI / Gemini, LangChain, SentenceTransformers
Database MongoDB Atlas (Vector Search)
Frontend (optional) Next.js, Tailwind CSS
Hosting Vercel (frontend), Render/Fly.io (backend)
Repo Cloning GitHub API / git

🧰 Features (Phase-wise)

✅ Phase 1: Project Setup & Planning

  • [x] Create a new GitHub repo (public, with a clear README)
  • [x] Initialize backend folder structure (Python/FastAPI or Node.js)
  • [x] Initialize frontend folder structure (Next.js, optional)
  • [x] Add .gitignore, LICENSE, basic README
  • [x] Set up MongoDB Atlas cluster with Vector Search enabled
  • [x] Set up OpenAI or Gemini API key access

✅ Phase 2: Repository Cloning & Parsing

  • [x] Create CLI or backend script to clone a public GitHub repo using GitPython or subprocess
  • [x] Recursively scan cloned repo for .js, .ts, .py, .java, .go, etc. files
  • [x] Filter out node_modules, .git, __pycache__, build, etc.
  • [x] For each file:
  • [x] Read file content
  • [x] Store file path and language
  • [x] Split into code chunks (use LangChain TextSplitter or custom logic)
  • [x] Associate each chunk with metadata (filepath, repo name, lang, etc.)
  • [x] Save chunks locally in JSONL or memory for now

✅ Phase 3: Embeddings + MongoDB Vector Search Integration

  • [x] Choose embedding model (OpenAI, HuggingFace, Gemini)
  • [x] Generate embeddings for each code chunk
  • [x] Connect to MongoDB Atlas (Python: pymongo + pymongo.vector_search)
  • [x] Create schema:
  • [x] Fields: content, filepath, language, embedding (vector), repo name, etc.
  • [x] Insert chunk docs into MongoDB collection
  • [x] Create vector index on the embedding field
  • [x] Test vector search: query for “auth middleware” or “database connection”

✅ Phase 4: AI Search + Summary Endpoint

  • [x] Create FastAPI or Express route: POST /query
  • [x] Input: natural language question + repo ID
  • [x] Perform vector search using MongoDB Atlas
  • [x] Return top-k chunks as context
  • [x] Pass context + query to LLM to generate a summary/answer
  • [x] Return final answer with citations (file paths)
  • [x] Test endpoint with: “How is login handled?”, “Where is JWT used?”

✅ Phase 5: Optional Frontend

  • [x] Scaffold a basic Next.js frontend
  • [x] Build form: Enter GitHub URL → triggers backend clone & index
  • [x] Build query input: Ask a question about the code
  • [x] Display AI result with file references
  • [x] List extracted endpoints, env vars, tech stack
  • [x] Add loading states and error handling
  • [x] Deploy frontend to Vercel (or keep CLI-only)

✅ Phase 6: Polish, Test, and Submit

  • [x] Write clean README with install/setup/run instructions
  • [x] Add screenshots or Loom demo
  • [x] Record 3-min demo video (host on YouTube/Vimeo)
  • [x] Host backend (e.g., Render, Railway, Google Cloud Run)
  • [x] Host frontend (e.g., Vercel)
  • [x] Add project + repo URLs to Devpost
  • [x] Submit before June 17 @ 4:00 PM CDT

📦 Local Development

Backend (FastAPI)

cd backend
python -m venv venv
source venv/bin/activate  # or .\venv\Scripts\activate on Windows
pip install -r requirements.txt
uvicorn main:app --reload

🌐 Frontend (Next.js)

cd frontend
npm install
npm run dev

🌍 Deployment

  • Frontend: Vercel ungithub.vercel.app
  • Backend: Render
  • Database: MongoDB Atlas with Vector Search enabled

🧠 Inspiration

Most devs struggle with onboarding into large codebases. We want to make codebases self-explanatory - like a smart dev mentor embedded into the repo itself.

🤝 Contributing

PRs are welcome! Feel free to fork this repo and open issues for bugs or features.

📄 License

MIT License

🙌 Built For

AI in Action Hackathon (Google x MongoDB x GitLab)

Share this project:

Updates