📘 Explainable PDF Question Answering (RAG App)

An Explainable Retrieval-Augmented Generation (RAG) system that allows users to upload a PDF, ask questions about its content, and receive:

• 📖 A grounded answer
• 🧠 An explanation of why specific context was used
• 📄 Source pages with relevance scores

Built using Gradio, LangChain, ChromaDB, and Hugging Face models.

---

🚀 Features

• 📂 Upload and index any PDF
• 🔍 Semantic search using embeddings (MiniLM)
• 🤖 Question answering using FLAN-T5
• 🧠 Explainability layer: why this answer?
• 📄 Source attribution with relevance scores
• 💻 Runs fully on CPU (Colab / low-resource friendly)

---

🧠 Architecture (High Level)

1. PDF Loader
   Uses PyPDFLoader to extract text from PDFs.

2. Text Chunking
   Splits text into overlapping chunks using RecursiveCharacterTextSplitter.

3. Embeddings

   • Model: sentence-transformers/all-MiniLM-L6-v2
   • Normalized embeddings for accurate similarity scoring.

4. Vector Store

   • ChromaDB for fast semantic retrieval
   • Persistent storage

5. LLM

   • google/flan-t5-base
   • Used twice:
     • Answer generation
     • Explanation generation

6. UI

   • Built with Gradio Blocks
   • Simple and interactive interface

---

🛠 Tech Stack

• Python
• Gradio
• LangChain
• ChromaDB
• Hugging Face Transformers
• PyTorch

---

📦 Installation

pip install gradio torch transformers langchain langchain-community chromadb sentence-transformers pypdf