🔍 Code Buddy: An AI-Powered Documentation Query Assistant

Overview

Code Buddy is an AI-driven assistant designed to answer technical questions using the latest documentation from various technologies. It leverages RAG (Retrieval-Augmented Generation) architecture combined with LangGraph, LangChain, and Pinecone to provide accurate, up-to-date responses by retrieving relevant documents and generating informed answers in real time.

This project was built during a hackathon with no prior experience in LangGraph. Everything was self-learned through documentation on the fly—demonstrating adaptability and a strong learning curve.

---

🔧 Features

• LangGraph & LangChain Integration: Modular, agentic workflow using state graphs and tools.
• Document Ingestion: Retrieves and stores documentation in a vector store using Pinecone.
• Search Augmentation: Queries search engines (via Tavily) to fetch recent, real-time content.
• Contextual Q&A: Answers user queries with context from retrieved documents.
• Powered by Cohere LLM: Generates coherent and accurate answers using Cohere's language models.

---

📦 Tech Stack

• LangGraph - Agentic framework for composing workflows
• LangChain - Framework for building LLM apps
• Pinecone - Vector store for semantic search
• Tavily - Real-time web search tool
• Cohere - LLM for generating answers
• Python - Implementation language
• Jupyter Notebook - Prototyping environment

---

🚀 How It Works

1. Document Retrieval:

   • Uses TavilySearchResults to fetch documentation.
   • Converts fetched pages into Document objects.

2. Vector Store Indexing:

   • Extracted documentation is embedded and indexed into Pinecone.

3. Query Handling:

   • User inputs a query.
   • Relevant documents are retrieved from Pinecone.
   • A prompt is dynamically generated.
   • Cohere LLM processes the prompt to generate an informed answer.

4. Agentic Execution:

   • LangGraph handles state transitions (START → RAG → END).
   • A tools-based conditional path determines whether to fetch or generate.

---

🧪 Example Use Case

User: "How does LangGraph manage state transitions?"

Code Buddy:

• Searches recent LangGraph docs using Tavily.
• Retrieves top pages and converts them into vector embeddings.
• Queries Pinecone for the most relevant chunks.
• Constructs a prompt and generates a concise, helpful answer using Cohere.

---

🛠️ Setup Instructions

1. Clone the repo

   git clone https://github.com/sharukh010/code_buddy_team_attention/
   cd code_buddy_team_attention

2. Install dependencies manually

   pip install -r requirements.txt

3. Set environment variables Create a .env file:

   TAVILY_API_KEY=your_tavily_key
   COHERE_API_KEY=your_cohere_key
   PINECONE_API_KEY=your_pinecone_key
   PINECONE_ENV=your_pinecone_env

4. Run the notebook Open code_buddy.ipynb in Jupyter and run all cells.

---

📈 Future Improvements

• Deploy as a web app using Streamlit or Gradio.
• Add support for multi-document summarization.
• Integrate chat history for contextual follow-ups.
• Enable fine-tuned LLM integration.

---

📄 License

MIT License. Feel free to fork, contribute, and build upon!