v2.0.0: Standalone tutorials, AI agents and MCP servers

A complete rewrite of the InterSystems FHIR/AI Hackathon Kit. It still covers similar ground to V1, but is restructured into standalone tutorials by data type. A Docker setup brings up InterSystems IRIS for Health with a FHIR server.

What it covers

Tutorials are Jupyter notebooks (with markdown versions) that you can run in order:

• Key concepts and setup: short introductions to IRIS, FHIR, Docker and RAG, plus environment setup
• Tabular data: load CSV data into IRIS SQL tables with Python, then query it
• FHIR data: load FHIR bundles and resources over REST, query FHIR from Python, and explore the FHIR specification with Swagger
• Vector search: chunk PDFs, embed them with OpenAI, store them in an IRIS vector column and search with SQL
• AI agents: build LangChain agents with tools that access the IRIS data
• MCP servers: wrap data access as MCP servers (FastMCP over HTTP, and stdio)
• Extras: create synthetic FHIR data, set up a FHIR server in 5 minutes, and an integrated machine learning quickstart

Setup scripts are included that recreate the tables earlier tutorials build, so you can start from a later tutorial.

What changed since V1

• Tutorials are now standalone rather than building one chatbot end to end
• New coverage of AI agents and MCP servers in Python
• OpenAI models replace local Ollama models, which were slow and computationally heavy
• Automated tests that run the notebooks against the Docker container
• Agent skills for the topics covered

The V1 patient chatbot with Ollama is still available as release v1.0.0 and the V1 branch.

Requirements

• Python
• Docker
• An OpenAI API key with some credit. The README estimates a full run of the tutorials costs less than $0.01.

See the README for setup steps.