AI-Powered Document Theme Chatbot

Published:

Project Overview

AI-Powered Document Theme ChatbotOpenAI GPT-4, LangChain, OCR, FAISS, Streamlit, FastAPI

This project solved the challenge of extracting and synthesizing information from a large, unstructured corpus of 75+ scanned PDFs and documents. I built an end-to-end RAG (Retrieval-Augmented Generation) system to enable sentence-level querying and theme-based summarization, drastically improving information retrieval efficiency.

Technology Stack

  • Generative AI: OpenAI GPT-4
  • Orchestration/Frameworks: LangChain, FAISS (Vector Store)
  • Deployment/Backend: Streamlit (Frontend), FastAPI (Backend API)
  • Data Pre-processing: OCR (Optical Character Recognition)

Key Achievements

  • Built a chatbot to query 75+ PDFs/scans with sentence-level citations using FAISS vector search.
  • Extracted text (incl. OCR) and enabled theme-based summarization across 1000+ paragraphs via LangChain.
  • Achieved 85% relevance in multi-doc answers during internal testing across 30+ queries.
  • Deployed a user-friendly web app with Streamlit and FastAPI, processing queries in <2.5 sec on average.