GitHub Codebase RAG Assistant

A Retrieval-Augmented Generation (RAG) application that lets you ask natural language questions about any GitHub repository. The system clones a repository, indexes the code files, generates embeddings, stores them in a Chroma vector database, and answers questions using an LLM grounded in the actual source code.

---

Architecture

User Question
     │
     ▼
┌─────────────┐       ┌────────────┐      ┌──────────────┐
│  Streamlit  │────▶ │  Retriever  │────▶│ Chroma Vector│
│     UI      │       │  (top-k=4) │      │    Store     │
└─────────────┘       └────────────┘      └──────────────┘
     │                                        ▲
     ▼                                        │
┌────────────┐                           ┌─────┴───────┐
│  RAG Chain │                           │  Ingestion  │
│  (Groq /   │                           │  Pipeline   │
│  LLaMA 3)  │                           └─────────────┘
└────────────┘                                ▲
     │                                        │
     ▼                                  ┌─────┴───────┐
  Answer +                              │  Repo Clone │
  Sources                               │  + Chunking │
                                        └─────────────┘

Pipeline

1. Clone — The target GitHub repository is cloned locally using GitPython.
2. Load — All supported source files (.py, .js, .ts, .md, .java, .cpp) are read.
3. Structure — A textual tree of the repository layout is generated.
4. Chunk — Files are split into overlapping 1 000-character chunks using LangChain's RecursiveCharacterTextSplitter.
5. Embed & Store — Chunks are embedded with HuggingFace embeddings (all-MiniLM-L6-v2, runs locally) and persisted in a Chroma vector database.
6. Retrieve — At query time, the 4 most similar chunks are retrieved.
7. Generate — Retrieved context is injected into a prompt and sent to Groq's llama-3.3-70b-versatile for a grounded answer.

---

Project Structure

├── src/
│   ├── repo_loader.py   # Clone repo & load files
│   ├── chunker.py        # Split documents into chunks
│   ├── ingest.py         # Full ingestion pipeline
│   ├── retriever.py      # Vector store retriever
│   ├── rag_chain.py      # LLM-powered RAG chain
│   └── utils.py          # Config, constants, helpers
│
├── vectorstore/           # Persisted Chroma database (auto-generated)
├── data/                  # Cloned repositories (auto-generated)
│
├── app.py                 # Streamlit UI entry point
├── requirements.txt
├── .env.example
└── README.md

---

Installation

1. Clone this project

git clone https://github.com/your-username/codebase-rag-assistant.git
cd codebase-rag-assistant

2. Create a virtual environment

python -m venv venv
# Windows
venv\Scripts\activate
# macOS / Linux
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables

Copy the example file and add your Groq API key:

cp .env.example .env

Edit .env:

GROQ_API_KEY=gsk_...

---

Usage

Start the application:

streamlit run app.py

1. Enter a GitHub repository URL in the sidebar (e.g. https://github.com/pallets/flask).
2. Click Index Repository — the repo will be cloned and indexed.
3. Type a question in the main area and click Get Answer.

---

Example Queries

Question	Sample Answer
Where is authentication implemented?	Authentication logic is implemented in src/auth.py where the login_user function verifies credentials against the database. Source: src/auth.py
Explain the structure of this repository.	The repository is organized into … Source: REPO_STRUCTURE
How does the database connection work?	Database connections are managed in src/database.py using a connection pool initialized in the connect() function. Source: src/database.py
What does the main script do?	The main entry point app.py starts a Flask server … Source: app.py

---

Tech Stack

• Python
• LangChain — orchestration, chunking, prompt management
• Groq API — LLM (llama-3.3-70b-versatile)
• HuggingFace / Sentence-Transformers — local embeddings (all-MiniLM-L6-v2)
• Chroma — local vector database
• GitPython — repository cloning
• Streamlit — web interface
• python-dotenv — environment variable management