SynapseHealer: Building an Autonomous, Self-Healing Interoperability DLQ with InterSystems IRIS AI Hub

SynapseHealer addresses the common issue of "dead-letter" messages, where corrupted or malformed data transactions halt critical processes. It employs an AI agent, which is explicitly constrained to a specific toolset for searching verified repair instructions, validating proposed fixes, and assessing risk. This design ensures the agent "cannot improvise" and that all repairs undergo validation before any message is resubmitted.

The system prioritizes efficiency and cost-effectiveness by defaulting to "fast, cheap model[s]" like gemini-1.5-flash, suitable for its "structural, not creative" decision-making. Performance metrics, derived from 10 regression suites and seven specific chaos scenarios, indicate an average cost of $0.00018 per heal and an end-to-end processing time of approximately 145 milliseconds, achieving a reported "90% auto-heal rate."

Governance and observability are central to SynapseHealer's operation. Every AI inference is logged to a persistent table, detailing costs, latency, and decision paths. A live dashboard tracks cumulative expenses and auto-heal rates. Each repair generates a "governed three-step trace," providing real-time visibility into the process, including side-by-side payload comparisons and the agent's reasoning, with human review options for complex cases.