RAG Prompt Engineering Experiments

An interactive RAG-based chatbot that allows users to upload and query PDF documents. The system uses FAISS vector search for retrieval and a local TinyLlama model for answer generation. It supports multiple prompt strategies, real-time chat interaction, and document-grounded responses with source tracking.

Key Features

• PDF ingestion and parsing using PyPDFLoader
• Text chunking with overlap for better context retrieval
• Vector embeddings using sentence-transformers (MiniLM)
• FAISS vector database for fast similarity search
• Retrieval-Augmented Generation (RAG) pipeline
• Local LLM inference using TinyLlama (no API required)
• Multiple prompt strategies (basic, few-shot, guardrail)
• Interactive chat UI built with Streamlit
• Conversation memory using session state
• Response latency tracking
• Source document highlighting for transparency
• Expandable UI to inspect retrieved chunks

Tech Stack

Screenshots