RAG Document Search

A Retrieval Augmented Generation (RAG) system for semantic document search and Q&A over PDF files, powered by a LangGraph ReAct agent.

Live Demo: http://rag-alb-1726979633.us-east-1.elb.amazonaws.com/

🎯 Features

• Document Upload: Upload and index PDF files directly from the browser
• Semantic Search: FAISS vector store with OpenAI text-embedding-3-small
• ReAct Agent: LangGraph ReAct agent for intelligent multi-step Q&A
• Web Search Fallback: Tavily search when answers aren't found in documents
• Source Citations: Every answer includes page-level source references
• Cloud-Native: S3 for PDFs, Secrets Manager for API keys, ECS Fargate for hosting

🛠️ Tech Stack

• Backend: FastAPI + Python 3.12
• AI: LangGraph ReAct agent, OpenAI embeddings, FAISS vector store
• APIs: OpenAI, Tavily
• Frontend: Vanilla JavaScript, HTML, CSS
• Cloud: AWS ECR + ECS Fargate + ALB + S3 + Secrets Manager
• CI/CD: GitHub Actions

🚀 Quick Start

Prerequisites

• Python 3.12+
• API Keys:
  • OpenAI API key
  • Tavily API key

Installation

1. Clone the repository:

git clone https://github.com/francis-rf/RAG-document-qa.git
cd RAG-document-qa

2. Install dependencies:

pip install -r requirements.txt

3. Create .env file:

cp .env.example .env
# Edit .env with your API keys

4. Run the application:

uvicorn app:app --reload --port 8000

5. Open browser:

http://localhost:8000

🐳 Docker Deployment

Build and Run

docker build -t rag-document-search .
docker run -p 8000:8000 --env-file .env rag-document-search

☁️ AWS Deployment

Services Used

Service	Purpose
ECR	Container image registry
ECS Fargate	Serverless container hosting
Application Load Balancer	HTTP traffic routing
S3 (rag-documents-qa)	PDF document storage
Secrets Manager (rag_document)	API key storage
CloudWatch	Logs and monitoring
IAM	Task roles and permissions

Setup

1. Store API keys in AWS Secrets Manager under secret name rag_document
2. Upload PDFs to S3 bucket rag-documents-qa
3. Push Docker image to ECR
4. Deploy via ECS Fargate with an ALB pointing to port 8000

Live URL

The app is deployed and accessible at:

http://rag-alb-1726979633.us-east-1.elb.amazonaws.com/

⚙️ GitHub Actions CI/CD

Automated deployment is configured via .github/workflows/deploy.yml.

Workflow: Deploy to AWS ECS

On every push to main, the pipeline:

1. Checks out the code
2. Configures AWS credentials
3. Logs in to Amazon ECR
4. Builds & pushes the Docker image to ECR (tagged with commit SHA and latest)
5. Triggers a force new deployment on ECS

Required GitHub Secrets

Add the following secrets to your GitHub repository (Settings > Secrets > Actions):

Secret	Description
AWS_ACCESS_KEY_ID	IAM user access key
AWS_SECRET_ACCESS_KEY	IAM user secret key

Workflow Status

📁 Project Structure

RAG-document-qa/
├── app.py                      # FastAPI application
├── src/
│   ├── config/                 # Settings — AWS Secrets Manager + .env fallback
│   ├── document_ingestion/     # PDF loading and chunking
│   ├── vectorstore/            # FAISS vector store management
│   ├── nodes/                  # LangGraph retriever + ReAct agent nodes
│   ├── graph_builder/          # LangGraph workflow builder
│   ├── state/                  # State schema (TypedDict)
│   └── utils/                  # Logging
├── static/                     # Frontend
│   ├── index.html
│   ├── script.js
│   └── style.css
├── data/                       # PDF documents (local only — S3 on AWS)
├── vectorstore/                # FAISS index (local only)
├── .github/workflows/          # CI/CD
│   └── deploy.yml
├── Dockerfile
├── .dockerignore
└── requirements.txt

📡 API Endpoints

Method	Endpoint	Description
GET	/	Serves frontend
GET	/api	Health check
GET	/api/files	List PDFs (S3 or local)
POST	/api/upload	Upload a PDF file
POST	/api/load	Index documents into vector store
POST	/api/query	Query documents with a question

📸 Screenshots

RAG Document Search Interface

📄 License

MIT License