HelpmateAI: RAG-based QA System for Research Papers

HelpmateAI is a Retrieval-Augmented Generation (RAG) system designed to answer complex questions from long research documents using OpenAI LLMs.

🧠 Problem Statement

Can we build a smart QA assistant that reads and understands research papers/ policy documents to return precise, context-rich answers to domain-specific queries?

📄 Data

• This system was tested on a real-world life insurance policy document, available here: Principal-Sample-Life-Insurance-Policy.pdf
• Chunked and embedded using sentence transformers

⚙️ Approach

• Embedding Layer: Sentence Transformers + chunking strategy
• Vector DB: ChromaDB
• Search Layer: query embedding + reranking using cross-encoders
• Generation Layer: OpenAI GPT with few-shot prompt templates

🧪 Results

• 90%+ retrieval accuracy in top-3 matches (via reranking)
• Meaningful multi-sentence generated answers

📷 Sample Outputs

🔍 Search Layer

💬 Generation Layer

📄 Documentation

See the detailed process and challenges in HelpmateAI_RAG_Project_Documentation.pdf

🧰 Tech Stack

Python, LangChain, ChromaDB, OpenAI, Transformers, Scikit-learn