iris-ai-examples

Working AI Hub applications built on InterSystems IRIS — ready to run, domain-specific, and built for demonstration.

Each example ships with a Docker stack, seeded demo data, and a set of MCP tools you can drive from Claude Desktop, VS Code, or any MCP client. No configuration required beyond an API key.

Examples

Example	Domain	Tools	AI Hub APIs	Interop
careconnect-sdoh	Healthcare / SDoH	9	%AI.ToolSet, %AI.MCP.Service	BS → BP → BO production
kg-ticket-resolver	Support / Knowledge Mining	6	%AI.ToolSet, %AI.MCP.Service, %AI.Agent, %AI.Provider	—

careconnect-sdoh/

Healthcare SDoH Assessment Agent

A community health worker assistant that assesses Social Determinants of Health for patients and triggers follow-up workflows through IRIS Interoperability.

• 9 MCP tools: patient lookup, SDoH risk scoring across 5 USDHHS domains, care plan generation, follow-up workflow trigger, interop message tracing
• IRIS Interoperability production wired end-to-end (BusinessService → BusinessProcess → BusinessOperation)
• 3 pre-seeded demo patients covering diabetes/hypertension, CHF/depression, and prenatal care
• Shows how an agent can observe and trigger production workflows — not just query data

Best for demonstrating: %AI.ToolSet, %AI.MCP.Service, IRIS Interoperability integration with AI agents, Ens.Director, live message tracing

---

kg-ticket-resolver/

Support Ticket Knowledge Mining Agent

A support engineer assistant that mines a backlog of 276 synthetic EMR support tickets, scores data completeness, finds similar tickets via vector search, and drafts KB articles using a %AI.Agent running inside IRIS.

• 6 MCP tools: MDS completeness scoring, semantic vector search (IRIS VECTOR_COSINE), cluster analysis, AI-generated KB articles, wiki management with Graph_KG provenance
• %AI.Agent + %AI.Provider pattern: an agent running inside IRIS called via MCP
• IRIS native vector search (VECTOR(DOUBLE, 384) + VECTOR_COSINE) — no external vector DB
• Jupyter notebooks showing the same pipeline from a data science perspective
• Pre-existing wiki with documented knowledge gaps, augmented by the agent

Best for demonstrating: %AI.Agent + %AI.Provider, IRIS native vector search, MDS scoring, knowledge graph provenance, agentic KB synthesis, notebook-friendly architecture

---

Requirements

• InterSystems IRIS AI Hub community build 162+
  • Download: evaluation.intersystems.com/Eval/early-access/AIHub
• Docker + Docker Compose
• An MCP client: Claude Desktop, VS Code with Copilot, or any MCP-compatible tool
• OpenAI API key (for DraftKBArticle and %AI.Agent tools — other tools work without one)

AI Hub Concepts Covered

Concept	Where demonstrated
%AI.ToolSet — define tools in ObjectScript XData	Both examples
%AI.MCP.Service — expose a ToolSet via MCP endpoint	Both examples
iris-mcp-server — connect any MCP client to IRIS	Both examples
%AI.Agent — run an LLM agent loop inside IRIS	kg-ticket-resolver: DraftKBArticle
%AI.Provider — configure LLM backends	kg-ticket-resolver: DraftKBArticle
IRIS Interoperability + AI — trigger BS/BP/BO from an agent tool	careconnect-sdoh
IRIS native vector search — VECTOR type + VECTOR_COSINE	kg-ticket-resolver: FindSimilarTickets
Graph_KG provenance — record agent actions as graph edges	kg-ticket-resolver: PublishKBArticle
Demo data seeding — idempotent %Persistent table population	Both examples
MCP sidecar pattern — iris-mcp-server alongside IRIS in Docker Compose	Both examples

Running an Example

Each example is self-contained. From any example directory:

cd <example>/docker
docker compose up -d

Then connect your MCP client to the running server. See each example's README for the exact client config.

Related

• ready-hackathon-dev-template — minimal starter if you're building something new
• iris-vector-graph — graph + vector engine for more advanced RAG patterns
• ready2026-knowledge-graph-demo — standalone notebook version of the ticket mining pipeline