document-chatbot-rag

A RAG chatbot for querying your own documents using open-source models. Upload documents, ask questions, get answers based on your content using vector search and local LLMs.

Quick Start

Prerequisites: Docker and Docker Compose only (no Python installation needed)

Start the System

./start-services.sh

This single command will:

• Start Qdrant vector database service
• Start Ollama LLM service
• Install Python dependencies automatically
• Download llama3.2 model
• Launch interactive chatbot

Stop the System

./stop-services.sh

Add Your Documents

Place your text or PDF files in the data/documents/ folder before starting the system.

Example Interaction

You: What is artificial intelligence?
Bot: Based on the context, Artificial Intelligence (AI) is a broad field 
     that encompasses machine learning, deep learning, and natural language 
     processing, aiming to create intelligent systems that can perform 
     tasks typically requiring human intelligence.

No local Python/pip installation required - everything runs in Docker containers.

How it works

1. The user uploads a text document -> "Machine learning helps computers learn from data"
2. This document is preprocessed (chunking) -> chunks = ["Machine learning helps", "computers learn from data"]
3. Chunks are embedded using transformers (SentenceTransformer) resulting in sentence vectors -> [[0.1, 0.8, ...], [0.2, 0.7, ...]]
4. These vectors are stored in a Vector DB (Qdrant) labeled properly with the text chunk they represent -> {vector: [0.1, 0.8, ...], text: "Machine learning helps"}
5. The user asks something -> "What is ML?"
6. User input is embedded and the vector is stored in the Vector DB
7. Similarity search is done to get the best result according to a score -> "Machine learning helps..." (score: 0.85)

Tech stack

Tool	Purpose
LangChain	RAG orchestration and chatbot pipeline
Qdrant	Vector database for semantic search
Sentence Transformers / HuggingFace	Embeddings generation
Ollama	Local open-source models
FastAPI	Main Framework
Docker	Run everything on a container

MVP Flow

1. Initialization (main.py):
   • Load hardcoded document from data/documents/
   • Process and save to vector store if not exists
2. Processing (document_processor.py):
   • Extract text from PDF
   • Split into chunks with overlap
3. Embeddings (embeddings.py):
   • Use local Sentence Transformers
   • Generate embeddings for chunks
4. Vector Store (vector_store.py):
   • Connect to Qdrant (Docker)
   • Store and search chunks
5. Chatbot (chatbot.py):
   • Receive user question
   • Retrieve relevant chunks
   • Generate response with Ollama