Agentic RAG with LangGraph & ChromaDB 🦜🕸️

A sophisticated, stateful Agentic RAG (Retrieval-Augmented Generation) system designed for intelligent document interaction. This agent doesn't just retrieve; it reasons about the necessity of retrieval and maintains a coherent conversation state across multiple turns.

🌟 Key Features

• 🧠 Agentic Reasoning: Powered by LangGraph, the system employs a cyclic graph-based workflow where the LLM autonomously decides whether to query the knowledge base or respond directly.
• 🔄 Stateful Persistence: Implements MemorySaver to preserve chat history, ensuring contextually aware responses within a single user session.
• ⚡ High-Performance Retrieval: Utilizes ChromaDB for local vector storage and OpenAI's text-embedding-3-small for precision-grade semantic search.
• 📄 Seamless Ingestion: Automated PDF processing, chunking, and indexing for immediate "chat-with-your-data" capabilities.

🛠️ Tech Stack

• Orchestration: LangGraph (StateGraph, Nodes, Edges)
• Framework: LangChain
• Vector Database: ChromaDB
• LLM: OpenAI GPT models (Optimized for GPT-4o / GPT-5)
• Embeddings: OpenAI text-embedding-3-small

📂 Architecture

.
├── chroma_db/                  # Local Vector Store persistence
├── your_document.pdf           # Target knowledge source
├── RAG_Agent.py                # Core Agentic logic & Graph definition
├── .env                        # Environment configuration
└── requirements.txt            # Project dependencies

🚀 Getting Started

1. Installation

git clone https://github.com/savinoo/RAG_Agent.git
cd RAG_Agent
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Configuration

Create a .env file:

OPENAI_API_KEY=your_key_here

3. Execution

Place your PDF in the root directory and run:

python RAG_Agent.py

🤖 How it Works

1. Ingestion: The script checks for an existing ChromaDB instance. If absent, it parses the PDF into semantic chunks and generates embeddings.
2. Decision Loop: When a query is received, the agent enters a LangGraph node.
3. Tool Call: If the agent determines retrieval is needed, it calls the ChromaDB tool.
4. Final Response: The agent synthesizes the retrieved context and conversation history into a final answer.

---

Developed by Lucas Lorenzo Savino