📄 Local PDF RAG Chatbot

An end-to-end Retrieval-Augmented Generation (RAG) pipeline built entirely with open-source tools. This application allows users to upload PDF documents and ask context-specific questions. All text extraction, embedding, and generation occurs 100% locally on-device, ensuring zero API costs and complete data privacy.

🚀 Pipeline Architecture

This project strictly follows the standard RAG architecture:

1. Data Ingestion: PyPDFLoader extracts text from the uploaded PDF.
2. Chunking: RecursiveCharacterTextSplitter divides the text into manageable 1000-character chunks with a 200-character overlap to preserve semantic context.
3. Embedding: Chunks are vectorized using the local nomic-embed-text model via Ollama.
4. Vector Storage: Vectors are stored locally in a Chroma database for rapid similarity search.
5. Retrieval & Generation: User queries are embedded, matched against the vector store using cosine similarity, and passed to a local Llama 3 model alongside the retrieved context to generate deterministic, grounded answers.

🛠️ Tech Stack

• Language: Python
• Orchestration: LangChain
• LLM Engine: Ollama (Llama 3 for generation, Nomic for embeddings)
• Vector Database: ChromaDB
• User Interface: Gradio

⚙️ Local Setup Instructions

Prerequisites

You must have Ollama installed on your machine to run the local models.

Once Ollama is installed, pull the required models via your terminal:

ollama run llama3
ollama pull nomic-embed-text