sqlgen-nb

Write PostgreSQL query from natural language using on-prem gen AI -- with IBM Granite Code LLM, Ollama on Juypter Notebook.

Inspired by a use case mentioned by Anthropic -- but is a local implementation with a minimal setup.

Installation

Jupyter Notebook https://marketplace.visualstudio.com/items?itemName=ms-toolsai.jupyter

LangChain https://python.langchain.com/docs/how_to/installation/

Ollama https://ollama.com/download

Pull IBM Granite Code model in Terminal:

ollama pull granite-code:8b-instruct 

Usage

1. Open sqlgen.ipynb
2. Enter your database schema in schema =
3. Enter your question in user_prompt =
4. Click Run all to run all the cells in the Juypter Notebook
5. Scroll to the bottom for the SQL query generated

Challenges & limitations

Gen AI is non-deterministic. With the same user prompt, it generates slightly different completions every time. Verify before use.

What's next?

As this involves interacting LLM with database. The codes could be reused in database agent in multiagent workflow.