---

ENHANCED PDF RAG CHATBOT: A RETRIEVAL-AUGMENTED GENERATION SYSTEM USING LANGCHAIN AND FAISS

📄 PDF Knowledge Assistant (RAG Chatbot)

A Streamlit-based PDF RAG chatbot where users can upload one or multiple PDFs, build a vector index, and ask questions to get grounded answers with source/page references.

Demo: https://drive.google.com/file/d/1FCID0_AZsBk-GosD5JzA1PF705j5Ou7M/view?usp=sharing

🚀 What it does:

Upload text-based PDFs (selectable text), Builds embeddings using SentenceTransformers, Stores and searches using FAISS, Answers questions using Gemini API, Shows top source chunks + page numbers, Warns users if the PDF looks scanned/image-based (OCR not enabled).

🧠 Why I built this:

To create a practical “chat with your PDFs” assistant that is fast, simple to use, and avoids hallucinations by answering only from retrieved document context.

🛠 Tech Stack: Python, Streamlit (UI), FAISS (vector search), SentenceTransformers (embeddings), LangChain Google GenAI (Gemini)

📦 Setup (Local)

1. Install dependencies pip install -r requirements.txt
2. Add your Gemini API key Create: .streamlit/secrets.toml GOOGLE_API_KEY = "YOUR_KEY_HERE"
3. Run the app streamlit run app.py

✅ Notes / Limitations

• Works best with text-readable PDFs
• Scanned PDFs may not work well because OCR is not enabled
• Similarity threshold and Top-K can be tuned from the sidebar[Best: Top-K = 5, Threshold = 0.15/0.16/0.20]

📌 Features I’m proud of

• Clean UI + chat history + export chat
• Confidence checks to avoid random answers
• Source citations (file name + page + similarity score)

Contact me for the app link Email: mahboobunnisa885@gmail.com