ASMR: Agentic AI Generates Schemas for Ship Maintenance Reports

The challenge of automatically generating structured schemas from unstructured historical data, particularly in specialized domains like ship maintenance, is significant. To address this, a new agentic AI framework called ASMR (Agentic Schema Generation for Ship Maintenance Report Writing) has been developed. ASMR operates with two specialized AI agents. The first, a Field Generation Agent, extracts semantic concepts from narrative reports and proposes candidate schema fields by adaptively clustering information at multiple granularities. The second agent, a Structural Optimizer Agent, then uses reinforcement learning to identify the most compact, informative, and non-redundant schema representations from these candidates. The resulting schemas are designed to guide human report authors, helping them produce more complete, consistent, and actionable maintenance and operational reports. Preliminary results indicate the promise of this approach, while also highlighting ongoing research challenges at the intersection of data management, agentic AI, and human-centered AI design.