AI Codebase Assistant

AI Codebase Assistant is a Retrieval-Augmented Generation (RAG) application that helps developers understand unfamiliar codebases by answering natural language questions about uploaded project files.

Users can upload a project as a ZIP file, and the application indexes the source code using vector embeddings to generate context-aware responses powered by Google Gemini.

---

Live Demo

https://ai-codebase-assistant-uu6k.onrender.com/

Note: The application is hosted on Render's free tier. The first request may take 1–2 minutes while the server starts.

---

Features

• Upload an entire project as a ZIP file
• Automatic extraction and processing of source code
• Semantic search using FAISS vector database
• AI-powered question answering using Google Gemini
• Intelligent question routing
  • Answers metadata queries locally
  • Uses RAG for conceptual code understanding
• Project overview generation
• Programming language detection
• Project structure explorer
• Referenced source file viewer with syntax highlighting
• Conversation history
• Confidence score for retrieved results

---

Tech Stack

Frontend

• Streamlit

Backend

• Python

AI & Machine Learning

• Google Gemini
• LangChain
• FAISS
• Sentence Transformers

Libraries

• PyPDF
• tiktoken
• python-dotenv

---

Project Structure

AI-Codebase-Assistant/
│
├── app.py
├── config.py
├── requirements.txt
├── components/
├── services/
├── utils/
├── prompts/
├── assets/
└── uploads/

---

Installation

Clone the repository:

git clone https://github.com/Mahelaka29/AI-Codebase-Assistant.git

Move into the project directory:

cd AI-Codebase-Assistant

Install the required packages:

pip install -r requirements.txt

Create a .env file:

GOOGLE_API_KEY=your_api_key

Run the application:

streamlit run app.py

---

How It Works

1. Upload a project as a ZIP file.
2. The application extracts the source code.
3. Source files are divided into semantic chunks.
4. Sentence Transformers generate vector embeddings.
5. Embeddings are stored in a FAISS vector database.
6. Relevant code is retrieved based on the user's query.
7. Google Gemini generates context-aware answers using the retrieved code.

Metadata-related questions are answered locally without invoking the language model, improving response speed and reducing unnecessary API usage.

---

Sample Questions

• What does this project do?
• Explain the authentication flow.
• Where is the database connection established?
• Describe the project architecture.
• How is user authentication implemented?
• Which files handle API requests?
• Explain this function.
• Where is the main business logic located?

---

Future Improvements

• GitHub repository import
• Multi-repository support
• Conversation memory
• Repository comparison
• Code dependency visualization
• Support for additional language models
• Docker deployment

---

Author

Mahelaka Bano

LinkedIn: https://linkedin.com/in/mahelakamansoori

GitHub: https://github.com/Mahelaka29