IRIS Production Guardian

IRIS Production Guardian

An AI-assisted observability copilot for InterSystems IRIS administrators and developers. It helps you observe an Interoperability Production, investigate incidents with evidence, and query documentation with traceable citations. Built entirely on InterSystems IRIS (ObjectScript first, IRIS as the only database) for the InterSystems PT Developer Community Programming Contest 2026.

Modules

• Production Monitor - live health of Services, Processes and Operations, queues, error counts and a timeline of events.
• AI Incident Investigator - given an incident, component and time window, it gathers real metrics, events and documentation and produces a hypothesis with evidence. Observation and hypothesis are kept separate; it is read-only, with no automated remediation.
• RAG Assistant - hybrid retrieval (IRIS Vector Search + iFind lexical search, fused with Reciprocal Rank Fusion), answers with citations, and calibrated abstention when the sources do not support an answer. Two generation providers are wired in (Gemini and Groq).

Also included

• Business Rules routing - Guardian.Rule.IncidentRoutingRule routes incidents by severity and is editable in the portal without recompiling.
• Public API access - a second, independent Production polling a real public API, with a cache and unavailable-fallback pattern that never invents data.
• Failure and recovery demo - a controlled failure scenario that shows the Monitor, Investigator and RAG working together.

Built with AI, documented honestly

The whole project was developed with AI coding assistance. The prompts, the mistakes the AI made and how each was caught and corrected are documented in the accompanying Developer Community article, including how hallucinations were mitigated.

Getting started

See the README for step-by-step instructions: run IRIS Community Edition in Docker, load the classes, start the Production, configure your AI provider key, ingest the RAG corpus and reproduce the failure and recovery demo.

Author: Sérgio Fernandes de Sousa Quinta (solo project).