🗂️ Repo Whisperer

Your friendly AI assistant for understanding GitHub repositories — just paste a repo URL, select a file, and ask questions!

---

🌍 Live Deployment: https://repo-whisperer.vercel.app
🫙 Backend: Start the backend Instance

🕒 Note: The backend may take a few seconds to wake up if it's been idle (Render cold start).

---

🚀 Features

• 🌐 Paste any public GitHub repo URL to load its file structure.
• 🗃️ View a TreeView of all repository files.
• 🧠 Select a file, ask a question — get context-aware answers powered by Cohere embeddings and Gemini/GPT.
• 📚 Answers are formatted with markdown + code blocks and are easy to read.
• ⚡ Instant feedback using toast notifications.
• 📦 Uses semantic chunking, vector similarity, and LLMs.

---

📸 Example Output

---

🛠️ Tech Stack

Frontend

• React + Vite
• Material UI (MUI)
• Framer Motion
• React Markdown for rich formatting
• React Syntax Highlighter for code blocks
• React Virtualized Autocomplete for performance
• Toastify for notifications

Backend

• FastAPI for serving endpoints
• Cohere API for vector embedding & chat
• Gemini API for answer generation
• Scikit-learn for cosine similarity
• Requests, dotenv, NumPy, etc.

---

🧪 How It Works

1. User enters GitHub repo URL → Frontend sends it to backend (/fetch-files)
2. Backend:
   • Extracts repo tree using PyGitHub or GitHub API
   • Returns list of files and tree structure
3. User selects a file & asks a question
4. Backend:
   • Fetches raw file content from GitHub
   • Splits code into chunks + embeds using Cohere
   • Embeddings saved to data/ folder
   • Finds most relevant chunks using cosine similarity
   • Sends top chunks + user question to Gemini/GPT
   • Returns answer

---

📂 Folder Structure

Repo-Whisperer/
│
├── frontend/                  # React frontend (UI)
│   └── src/
│       └── App.jsx           # Core component (URL input, QnA interface)
│
├── backend/                  # FastAPI backend
│   ├── main.py               # API routes (/fetch-files, /ask)
│   ├── ingest/
│   │   ├── fetch_repo.py     # GitHub file listing
│   │   └── chunk_and_embed.py # Chunking + embedding logic
│   └── query/
│       └── ask_question.py   # Semantic search + LLM generation
│
├── data/                     # Stores JSON embeddings per file
└── README.md

🧠 Powered By

• Cohere – for embeddings, conversational model and advanced LLM responses
• GitHub REST API – for repo access
• FastAPI – for backend API

💡 Future Enhancements

• ✅ GitHub Auth (for private repos)
• 🔍 File-level search
• 🧪 Explain function/class on hover
• 🌍 Multi-language support

🙌 Author

Made with ❤️ by Sumedh