Inspiration
IBM Maximo EAM serves 1,000+ critical infrastructure organizations — utilities, hospitals, governments, oil and gas. Existing AI tools hallucinate work order numbers and asset IDs. In a power plant, that is not acceptable. MAXIMO SENTINEL was built to solve this: a governance-first multi-agent system that refuses to answer when evidence is insufficient.
What it does
MAXIMO SENTINEL is a 4-agent enterprise asset intelligence system:
- Orchestrator Agent — applies 60 governance rules before every response
- KB Retrieval Agent — semantic search over 4,912 real Maximo procedures via Elastic MCP
- Compliance Agent — governance scoring and ITIL rule mapping
- Work Order Triage Agent — PM overrun detection and priority routing
The agent refuses to fabricate asset IDs, work order numbers, or compliance statuses. Every response includes a source citation and confidence score.
How we built it
Built with Google Cloud Agent Builder (ADK 2.0), Gemini 2.5 Pro, Elastic Cloud MCP Server, and deployed on Google Cloud Run. The governance layer enforces 60 rules derived from 17 years of enterprise asset management experience.
Challenges
Building a system that knows what it does not know — and refuses to guess — required rethinking agent architecture from first principles. The governance refusal pattern is the core innovation.
What we learned
Governance-first agent design produces more trustworthy outputs than capability-first design. The recursive self-verification loop — where the agent reads its own output before delivery — eliminates hallucination structurally rather than probabilistically.
Log in or sign up for Devpost to join the conversation.