AI Playground (ai_pg)

This project is a playground for testing the integration of LLM capabilities into CKAN. It utilizes Llama Index, Ollama, and FastMCP to demonstrate how to generate metadata from data files.

Prerequisites

Before setting up the project, you must ensure that Ollama is installed and the required model is available.

1. Install Ollama

Download and install Ollama for your operating system.

2. Pull the Required Model

This project uses the qwen3:8b model. Once Ollama is installed, open your terminal and run:

ollama pull qwen3:8b

Installation

1. Clone this repository and navigate to the project directory.
2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate

3. Install the project:

   pip install -e .

Usage

This tool provides two main CLI commands to interact with the LLM and the MCP server.

1. Run the MCP Server

The MCP (Model Context Protocol) server needs to be running, as it provides tools for the agent.

ai_pg run-mcp

This command starts the local MCP server located in mcp_client/main.py.

2. Generate Metadata

This command queries the local Ollama LLM and the MCP server to generate metadata for a specific dataset file.

Note: The file you specify must exist in the data folder. You can also specify a file from your file system, but stick to the CSV format, as other formats are not yet supported.

ai_pg generate-metadata -f ./data/titanic.csv

RAG (Retrieval-Augmented Generation)

RAG related commands:

1. RAG Index documents

This one indexes all documents in the data folder to the postgres. See the config.py for the configuration of the database and embedding model.

2. RAG Index document

To index a specific document, run:

ai_pg rag index-document titanic.csv

This is a simple way to re-index a document if it has been updated.

3. RAG Query

To query the indexed documents, run:

ai_pg rag query-index "Who is Mr. Owen Harris Braund?"

Mr. Owen Harris Braund is a male passenger who was 22 years old and traveled in
the third class on the Titanic. He was listed as a crew member (indicated by the
value 1 in the third column), and his ticket class was A/5 21171, with a fare of
7.25. He embarked from Southampton (S).

Configuration

You can configure the Ollama model, MCP server and other things using environment variables. See the config.py file for default settings.