Astrology RAG System

A Retrieval-Augmented Generation (RAG) chatbot for Astrology knowledge using LangChain, ChromaDB, E5 Embeddings, MMR Retrieval, BGE Reranker and Qwen Local LLM.

Features

• Markdown document ingestion
• Header-based semantic chunking (#, ##, ###)
• Multilingual embeddings (E5)
• ChromaDB vector database
• MMR retrieval
• BGE reranker
• Local LLM inference using Ollama

Project Structure

src/ ├── ingestion/ │ └── loader.py ├── chunking/ │ └── chunker.py ├── embeddings/ │ └── embedder.py ├── vectorstore/ │ └── vector_store.py ├── retrieval/ │ ├── retriever.py │ └── reranker.py ├── llm/ │ └── llm_client.py └── prompts/ └── prompt_template.py

Pipeline

Markdown Files ↓ Loader ↓ Header Chunking ↓ E5 Embedding ↓ ChromaDB ↓ MMR Retrieval ↓ BGE Reranker ↓ Qwen3 ↓ Answer

Installation

pip install -r requirements.txt

Create Vector Database

python create_db.py

Run Chatbot

python main.py

Technologies

• LangChain
• ChromaDB
• HuggingFace Embeddings
• BAAI BGE Reranker
• Ollama
• Qwen3