📄 Serverless RAG Chatbot on AWS

A production-ready Serverless Retrieval-Augmented Generation (RAG) system built on AWS Lambda, API Gateway, Pinecone, and OpenAI. The system allows users to upload documents (PDF) and ask natural language questions via a ChatGPT-like web UI.

🚀 Key Features

• 🔍 Semantic Search (RAG) using vector embeddings
• 📚 PDF ingestion & chunking
• 🧠 LLM-based answer generation (OpenAI)
• ⚡ Serverless architecture (AWS Lambda + API Gateway)
• 🐳 Container-based Lambda (ECR)
• 🌐 ChatGPT-like Web UI (pure HTML/CSS/JS)
• 🔐 Secrets managed via GitHub Secrets
• 📦 Infrastructure as Code with Terraform
• 🔄 CI/CD with GitHub Actions

---

🏗️ High-Level Architecture

User (Browser)
   ↓
index.html (Chat UI)
   ↓  POST /ask
AWS API Gateway
   ↓
AWS Lambda (Query Handler)
   ├─ Retrieval → Pinecone Vector DB
   └─ Generation → OpenAI Chat Model
   ↓
Answer returned to UI

A separate Ingestion Lambda is used to process and embed documents.

---

📂 Project Structure

SERVERLESS_RAG_PROJECT
├── .github/
│   └── workflows/
│       └── deploy.yml          # CI/CD pipeline
├── infra/                      # Terraform IaC
│   ├── main.tf
│   ├── variables.tf
│   └── outputs.tf
├── src/
│   ├── common/
│   │   └── logger.py
│   ├── ingestion/
│   │   ├── handler.py          # Document ingestion Lambda
│   │   └── service.py
│   └── retrieval/
│       ├── handler.py          # Query Lambda
│       ├── search.py
│       └── generator.py
├── index.html                  # ChatGPT-like frontend
├── Dockerfile
├── requirements.txt
└── README.md

---

🧠 RAG Flow Explained

1️⃣ Ingestion Phase

• Upload PDF documents
• Extract text per page
• Split text into chunks
• Generate embeddings using OpenAI Embeddings
• Store vectors in Pinecone

2️⃣ Query Phase

• User submits a question
• Convert question to embedding
• Perform similarity search in Pinecone
• Retrieve top-K relevant chunks
• Inject context into prompt
• Generate final answer via OpenAI Chat Model

---

🛠️ Technology Stack

Layer	Technology
Frontend	HTML, CSS, Vanilla JS
API	AWS API Gateway
Compute	AWS Lambda (Container Image)
Container	Docker + Amazon ECR
Vector DB	Pinecone
Embeddings	OpenAI text-embedding-3-small
LLM	OpenAI Chat Models
IaC	Terraform
CI/CD	GitHub Actions

---

🔐 Environment Variables

Configured via GitHub Secrets and injected by Terraform:

OPENAI_API_KEY=
PINECONE_API_KEY=
PINECONE_INDEX_NAME=
PINECONE_NAMESPACE=default

⚠️ Changing env vars does NOT require rebuilding Docker images — only a redeploy.

---

🔄 CI/CD Pipeline (GitHub Actions)

Triggered on push to master:

1. Checkout code
2. Build Docker image
3. Push image to Amazon ECR
4. Run Terraform
5. Update Lambda functions

Image tagging strategy:

• latest
• ${GITHUB_SHA} (immutable)

---

🐳 Docker & Lambda

• Base image: public.ecr.aws/lambda/python:3.12
• Optimized for fast cold start
• No heavy ML libraries (no torch, no sentence-transformers)

Lambda handlers:

src.ingestion.handler.handler
src.retrieval.handler.handler

---

🌐 Frontend Usage

Open index.html in a browser and ask questions:

POST /ask
{
  "query": "What is the main content of ML.pdf?"
}

Features:

• ChatGPT-style UI
• User / Bot message alignment
• Loading indicator
• CORS enabled

---

💰 Cost Considerations

• 💵 Pay-as-you-go (OpenAI)
• Pinecone: free / starter tier suitable for small docs
• AWS Lambda: extremely low cost for light usage

Example:

• 1 PDF (~6 pages): < $0.01 embedding cost
• Typical query: fractions of a cent

---

✅ Production Best Practices Applied

• No secrets in Docker images
• Immutable image tags
• Environment-based configuration
• Stateless Lambdas
• Warm-start optimization

---

🚧 Future Improvements

• 🔐 Authentication (Cognito / JWT)
• 📎 Source citation in answers
• 🔄 Streaming responses
• 📊 Observability & metrics
• 🌍 Multi-language support

---

👤 Author

Built by Thanh – Backend Engineer

Focused on AWS, Serverless, and AI-powered systems.

---

⭐ Summary

This project demonstrates a clean, scalable, and production-ready RAG architecture using modern cloud-native and AI technologies.

If you are learning Serverless + GenAI, this is a solid real-world reference.