AegisFlow: A Multi-Agent Agentic AI Framework for Autonomous Remediation and Self-Healing in Fragile Data Ecosystems

01 ABSTRACT

The paper proposes AegisFlow, a multi-agent framework for autonomous data pipeline remediation, comprising Watchdog and Repair agents that use LLMs to automatically generate, test, and deploy code patches. The framework employs a parallel shadow patching execution model based on the MAPE-K loop to validate patches in digital twin environments. The authors report that across five failure scenarios, average MTTR dropped from 170 minutes to 3.2 minutes, with a 92% patch success rate, and claim it can free about 98% of data engineering on-call time.

02 KEY FINDINGS

1. The Watchdog agent collects runtime telemetry, while the Repair agent uses LLMs to automatically create, test, and deploy code patches.
2. It adopts a parallel shadow patching non-intrusive execution model based on the MAPE-K loop to generate and verify patches in digital twin environments.
3. Across five failure scenarios, it reports average MTTR reduced from 170 minutes to 3.2 minutes, a 92% patch success rate, including 96% for JSON schema changes, 98% for punctuation drift, and 85% for Shadow DOM.