AWS Customer Agreement RAG Assistant

Overview

This project is a Retrieval-Augmented Generation (RAG) system built on top of the AWS Customer Agreement document.

The application allows users to ask natural language questions about the agreement and receive grounded answers generated from relevant document sections.

The system combines semantic search using FAISS and Sentence Transformers with Large Language Model (LLM) generation using Groq.

Scope: the current build indexes a single document (the AWS Customer Agreement). Multi-document support and reranking are listed under Future Improvements.

---

Features

• PDF document parsing
• Text chunking and preprocessing
• Semantic embeddings using Sentence Transformers
• FAISS vector database for similarity search
• Context-aware answer generation using Groq LLM
• FastAPI backend
• SQLite query logging
• Analytics endpoint
• Streamlit user interface
• Swagger API documentation

---

Architecture

Document Flow:
AWS Customer Agreement PDF
    ↓
POST /ingest
    ↓
PDF Parser
    ↓
Chunking
    ↓
Embeddings (all-MiniLM-L6-v2)
    ↓
FAISS Vector Store

--------------------------------

Query Flow:
User Query
    ↓
POST /ask
    ↓
Retriever (Top-K)
    ↓
Groq LLM
    ↓
Generated Answer
    ↓
SQLite Logging

--------------------------------

Analytics Flow:
SQLite Logs
    ↓
GET /analytics
    ↓
Streamlit Dashboard

---

Tech Stack

Backend

• Python
• FastAPI
• SQLAlchemy
• SQLite

RAG Components

• Sentence Transformers (all-MiniLM-L6-v2)
• FAISS
• PyPDF

LLM

• Groq API

Frontend

• Streamlit

Setup Instructions

1. Clone Repository

git clone <repo-url>
cd vestaff-rag

2. Create Virtual Environment

python -m venv venv

Activate:

venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure Environment Variables

Create a .env file:

GROQ_API_KEY=your_api_key

5. Create Database

python -m backend.init_db

6. Generate Embeddings and FAISS Index

python test_ingest.py

7. Start FastAPI

uvicorn backend.main:app --reload

Swagger:

http://127.0.0.1:8000/docs

8. Start Streamlit Frontend

python -m streamlit run frontend/app.py

---

API Endpoints

GET /

Health check endpoint.

Returns a simple message confirming that the API is running.

---

POST /ingest

Processes and ingests the AWS Customer Agreement PDF.

This endpoint:

• Extracts text from the PDF
• Chunks the document into smaller sections
• Generates semantic embeddings using Sentence Transformers
• Builds and stores a FAISS vector index for retrieval

Returns:

• Success message
• Number of chunks created

---

POST /ask

Accepts a natural language query and executes the complete RAG pipeline.

Returns:

• Generated answer
• Retrieved source context
• Response latency

All interactions are automatically logged into the SQLite database for analytics.

---

GET /analytics

Returns usage statistics generated from SQL queries on the interaction logs.

Returns:

• Total queries
• Unanswered queries
• Average response latency
• Most frequently asked questions

Example Questions

• What are the responsibilities of users regarding security and backup?
• What is AWS indemnification policy?
• Can AWS modify the agreement?
• What are the payment obligations?

---

Analytics

The system logs every query in SQLite and tracks:

• Query count
• Answer success rate
• Average response latency

---

Future Improvements

• Hybrid Retrieval (BM25 + FAISS)
• Reranking Layer
• Multi-document support
• Conversation Memory
• Docker Deployment