RAG with TiDB and Ollama Tutorial

🎥 Video Walkthrough

You can watch the video version of this tutorial here:
▶ Watch on YouTube

Details:

This repository contains a step-by-step tutorial on how to build a local Retrieval-Augmented Generation (RAG) chatbot using:

• TiDB with the VECTOR data type
• Ollama for embeddings and LLM inference
• Python for ingestion and querying scripts

The tutorial shows how to:

• Convert TiDB documentation into a PDF and split it into chunks
• Store the text and embeddings in TiDB
• Query the database with semantic search
• Generate answers with Gemma 12B in a chatbot loop

⚠️ Note:

• This tutorial is intended as a learning project.
• For simplicity, it does not use TiFlash ANN indexes, but mentions them as an option for production.
• Running locally requires a GPU with ≥12GB VRAM or an Apple Silicon Mac.

---

Repository Contents

• tidb_docs_pdf.txt → link to an already generated TiDB documentation in pdf
• docs/index.html → styled HTML version for publishing
• load.py → script to split and embed documentation into TiDB
• chat.py → chatbot script that retrieves context and answers questions

---

🚀 Getting Started

1. Install TiUP and start TiDB Playground.
2. Install Ollama and pull the models:

   ollama pull nomic-embed-text:v1.5
   ollama pull gemma3:12b

3. Create a Python virtual environment and install requirements:

   python3 -m venv rag_tidb
   source rag_tidb/bin/activate
   pip install pymupdf pymysql ollama
   

4. Follow the tutorial in docs/index.html

Author Note

I’m a database engineer, not an AI expert. This tutorial was created with AI assistance as part of my own learning. There are probably better ways to implement RAG pipelines — feedback and improvements are very welcome!