CodeLens — AI Codebase Intelligence Engine

RAG-Powered Semantic Code Search & AI Chat for Any GitHub Repository

CodeLens lets you paste any public GitHub repo URL, indexes the entire codebase into a vector database, and then lets you ask questions in plain English — getting precise, context-aware answers grounded in the actual source code.

---

Features

Feature	Description
RAG Pipeline	Retrieval-Augmented Generation using FAISS vector search + Gemini LLM
File Explorer	Interactive file tree with extension-based icons and search filtering
AI Chat	Ask questions about any codebase in natural language
Conversation Mode	Multi-turn chat with sliding window context (last 10 messages)
Context Sources	See exactly which code chunks were used, with similarity scores
Auto Summary	Generates a 3-sentence codebase overview on indexing
Model Fallback	Automatically rotates through Gemini models if quota is hit

---

Architecture

User Question
     ↓
① EMBED the question (Gemini Embedding API)
     ↓
② IT retrieve's top-8 most similar code chunks (FAISS cosine similarity)
     ↓
③ AUGMENT the LLM prompt with retrieved chunks as context
     ↓
④ GENERATE answer using Gemini Flash
     ↓
Answer + Source Citations + Similarity Scores

---

Tech Stack

• Backend: Python, FastAPI, Uvicorn
• Vector DB: FAISS (Facebook AI Similarity Search)
• Embeddings: Google Gemini Embedding API (gemini-embedding-001)
• LLM: Google Gemini Flash (with automatic model fallback)
• Text Splitting: LangChain RecursiveCharacterTextSplitter
• Git: GitPython for repository cloning
• Frontend: Vanilla HTML/CSS/JS with Marked.js + Highlight.js

---

Quick Start

1. Clone the repo

git clone https://github.com/upadhyay74aman/Code_Lens.git
cd Code_Lens

2. Install dependencies

pip install -r requirements.txt

3. Get a Gemini API key

• Go to Google AI Studio
• Create a free API key

4. Create .env file

GEMINI_API_KEY=your_api_key_here

5. Run the server

python main.py

Or:

uvicorn main:app --port 8000

6. Open in browser

Navigate to http://localhost:8000 and start exploring codebases!

---

📡 API Endpoints

Method	Endpoint	Description
GET	/	Serves the frontend UI
POST	/api/index	Clone & index a GitHub repository
POST	/api/query	Single Q&A semantic search
POST	/api/chat	Multi-turn conversation mode
GET	/api/filetree	Get the indexed repo's file tree
GET	/api/status	Get indexing statistics

---

Project Structure

Code_Lens/
├── main.py           # FastAPI backend (RAG pipeline, FAISS, Gemini)
├── index.html        # Frontend UI (IDE-style dark theme)
├── requirements.txt  # Python dependencies
├── .env              # API key (not committed)
├── .gitignore        # Git ignore rules
└── README.md         # This file

---

How RAG Works in CodeLens

1. Indexing Phase: Repository files are split into ~1000 character chunks with 200 char overlap. Each chunk is embedded using Gemini's embedding model and stored in a FAISS Inner Product index.

2. Query Phase: Your question is embedded using the same model, then FAISS finds the top-8 most semantically similar code chunks via cosine similarity.

3. Generation Phase: The retrieved chunks are injected into a carefully crafted prompt, and Gemini Flash generates a grounded answer citing specific files and line numbers.

---

License

MIT License — feel free to use, modify, and distribute.