🚀 CodePilot AI

What is CodePilot AI?

CodePilot AI is an AI-powered developer assistant that helps engineers understand, search, and interact with source code using Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG).

Instead of relying only on keyword search, CodePilot AI understands the semantic meaning of code, allowing developers to ask natural language questions and receive accurate, context-aware answers grounded in the actual repository.

---

🌟 What Can CodePilot AI Do?

🔐 Secure User Authentication

• User registration and login using JWT authentication.
• Secure password hashing with BCrypt.
• Protected API endpoints.

🔗 GitHub Integration

• Connect GitHub accounts using OAuth 2.0.
• Access user repositories securely.
• Retrieve repository metadata and repository lists.

📥 Repository Cloning

• Clone public and authorized private GitHub repositories.
• Store repositories locally in a workspace for analysis.
• Manage multiple repositories for different users.

📂 Intelligent Repository Indexing

• Scan the repository structure recursively.
• Ignore unnecessary files and folders (.git, node_modules, venv, binaries, etc.).
• Detect programming languages based on file extensions.
• Build an index of all source files for efficient processing.

✂️ Semantic Code Chunking

• Parse source code using language-aware strategies.
• Create chunks based on: Classes, Functions, Large methods, Remaining module-level code.
• Preserve metadata such as: Repository, File path, Language, Class name, Function name, Line numbers.
• Split oversized classes and functions into multiple overlapping chunks while maintaining context.

🧠 Embedding Generation

• Convert every code chunk into a high-dimensional embedding using an AI embedding model.
• Represent the semantic meaning of code rather than simple keywords.

🗄️ Vector Database Storage

• Store embeddings in ChromaDB.
• Save metadata alongside vectors for filtering.
• Support efficient similarity search across thousands of code chunks.

🔎 Semantic Code Search

• Search repositories using natural language instead of keywords.
• Retrieve the most relevant code snippets based on semantic similarity.
• Filter search results by repository and metadata.
• Rank results by similarity score.

🤖 AI-Powered Code Assistant (RAG)

• Combine semantic search with an LLM to answer repository-specific questions.
• Generate grounded explanations using retrieved source code.
• Avoid hallucinations by providing answers based on actual repository content.
• Include references to the source files and functions used in the answer.

💬 Natural Language Repository Chat

Developers can ask questions such as:

• "How is JWT authentication implemented?"
• "Where is the database connection configured?"
• "Explain the GitHub OAuth flow."
• "Which service generates embeddings?"
• "How are repositories indexed?"
• "Where is SQLAlchemy initialized?"

The assistant retrieves the relevant code and generates a contextual explanation.

📚 Conversation Management

• Maintain conversation history.
• Support multi-turn interactions.
• Provide context-aware follow-up responses.

⚡ High Performance

• Dockerized deployment.
• PostgreSQL for structured data.
• Redis for caching/background tasks.
• ChromaDB for vector search.
• Optimized indexing and retrieval pipeline.

🛡️ Production-Ready Architecture

• Modular FastAPI architecture.
• Repository pattern & Dependency injection.
• SQLAlchemy ORM & Alembic migrations.
• Structured logging & Error handling.
• Environment-based configuration.
• RESTful APIs with OpenAPI/Swagger documentation.

---

🏗️ Project Architecture

User
 │
 ▼
JWT Authentication
 │
 ▼
GitHub OAuth
 │
 ▼
Clone Repository
 │
 ▼
Index Files
 │
 ▼
Semantic Chunking
 │
 ▼
Generate Embeddings
 │
 ▼
Store in ChromaDB
 │
 ▼
Semantic Search
 │
 ▼
LLM
 │
 ▼
AI Response

---

🛠️ Tech Stack

Backend

• FastAPI
• SQLAlchemy
• PostgreSQL
• Alembic
• Redis

AI

• Grok / OpenAI Compatible API
• ChromaDB
• Embedding Models
• RAG Pipeline

DevOps

• Docker
• Docker Compose

---

🚀 Installation & Setup

Clone the Repository

git clone https://github.com/<username>/codepilot-ai.git
cd codepilot-ai

Environment Variables

Create a .env file in the root directory:

DATABASE_URL=
SECRET_KEY=

GITHUB_CLIENT_ID=
GITHUB_CLIENT_SECRET=
GITHUB_CALLBACK_URL=

AI_API_KEY=
AI_MODEL=

CHROMA_DB_PATH=

Run with Docker (Recommended)

docker compose up -d

The API will be available at http://localhost:8000/docs

Manual Setup (Without Docker)

Create and activate a virtual environment:

# Windows
python -m venv venv
venv\Scripts\activate

# Linux/macOS
python -m venv venv
source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Run FastAPI locally:

uvicorn app.main:app --reload

---

💡 Example Workflow

1. Register an account and login.
2. Connect your GitHub account via OAuth.
3. Select and clone a target repository.
4. Index the source code.
5. Generate semantic chunk embeddings and store in ChromaDB.
6. Ask natural language questions about the codebase.
7. Receive accurate, grounded AI-powered answers.

---

🚀 Future Improvements

• Multi-language parsing
• Organization support
• Incremental indexing
• Webhooks
• CI/CD
• Monitoring
• Role-based access

---

📝 License

MIT License