Retrieval-Augmented-Workflow-using-Qdrant

A retrieval-augmented generation (RAG) workflow that transparently answers contract-related questions over the CUAD (Contract Understanding Atticus Dataset). This project leverages:

• LangGraph for explainable, multi-step agent workflows
• Qdrant as a high-performance vector database for semantic search
• OpenAI (GPT-4) for generative answering
• FastAPI (with built-in Swagger UI) to expose a REST API and a simple HTML form

The full tutorial is available as Part 6 of my series on Medium:

Implementing LLM Applications Using LangChain (Part 6): Retrieval-Augmented Agents with LangGraph & FastAPI

---

Table of Contents

1. Project Overview
2. Prerequisites
3. Installation
4. Directory Structure
5. Usage
6. Testing & Swagger UI
7. Contributing
8. License

---

Project Overview

This repository demonstrates a complete pipeline to build a production-ready RAG agent:

1. Indexing: Extracts contract clauses from the CUAD JSON and stores vector embeddings in Qdrant.
2. Agent Logic: Uses LangGraph to define a clear workflow—receive a question, retrieve relevant clauses, generate an answer via GPT-4.
3. API Exposure: Wraps the agent in a FastAPI server, providing both a JSON-based endpoint and a simple HTML form.

By following this repository, you’ll learn how to go from raw contract data to a fully functional LLM service capable of answering complex legal questions in real time.

---

Prerequisites

• Python 3.9+ (tested on 3.10/3.11/3.12)
• Docker (to run Qdrant locally)
• OpenAI API Key (with access to GPT-4 or GPT-3.5)
• Git (to clone this repo)

---

Installation

1. Clone the repository

   git clone https://github.com/Think-Round-Inc/ThinkxLife-CUAD-RAG.git
   cd ThinkxLife-CUAD-RAG

2. Create and activate a virtual environment

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

3. Install dependencies

   pip install --upgrade pip
   pip install fastapi uvicorn python-dotenv langchain langchain-openai langchain-qdrant langgraph qdrant-client

4. Create a .env file at the project root with your OpenAI API key:

   OPENAI_API_KEY=sk-<your_openai_key_here>

   Do not commit your .env file or share your key publicly.

---

Directory Structure

├── data/
│   └── cuad.json               # CUAD dataset (Contract Understanding Atticus)
├── build_index.py              # Extracts & indexes CUAD clauses into Qdrant
├── chatbot_core.py             # LangGraph agent logic (retrieval + LLM chain)
├── main.py                     # FastAPI server exposing `/ask` and HTML form
├── .env                        # (not checked in) environment variables (OpenAI key)
└── README.md                   # This file

---

Usage

Follow these steps to get the RAG agent up and running:

1. Configure Environment Variables

Ensure a file named .env exists at the project root containing:

OPENAI_API_KEY=sk-<your_openai_key_here>

2. Run Qdrant via Docker

Start a local Qdrant instance on port 6333:

docker run -p 6333:6333 qdrant/qdrant

Verify that Qdrant is running:

curl http://localhost:6333/collections
# Response: {"result":{"collections":[]}, ...}

3. Index CUAD Data

Populate Qdrant with contract clause embeddings:

python build_index.py

This step:

• Loads each clause’s text from data/cuad.json
• Embeds them via the OpenAI API
• Stores the resulting vectors in Qdrant under the collection named cuad_contracts

4. Start the FastAPI Server

Launch the API server:

python -m uvicorn main:app --reload

• The server runs on http://127.0.0.1:8000 by default
• The --reload flag restarts the server on code changes

---

Testing & Swagger UI

JSON Endpoint

1. Open http://127.0.0.1:8000/docs in your browser.
2. Use the POST /ask endpoint to submit JSON with a "question".
3. Review the generated answer in the Swagger UI response pane.

HTML Form

1. Open http://127.0.0.1:8000/ in your browser.
2. Enter a question and submit.
3. View the answer on the results page.

---

Contributing

Contributions are welcome! If you find bugs or have suggestions (e.g., multi-tool orchestration, guardrails integration, deployment scripts), please open an issue or submit a pull request.

---

License

This project is licensed under the MIT License. Feel free to reuse, modify, and redistribute under the MIT terms.

---