Context-Aware Multi-Agent Ai Framework FOR Autonomous Quality Engineering in Enterprise Banking Platforms

Abstract

Core-banking modernisation is one of the most risk-sensitive software programs a financial institution undertakes, and the quality-engineering function is where that risk is coordinated in practice. Present-day AI-assisted testing tools apply generative models as point solutions, without an orchestrating architecture that understands banking-specific regulatory context, reasons across an integrated core-banking ecosystem, or operates under the governance discipline that regulated environments require. This paper contributes a reference architecture, the Context-Aware Multi-Agent AI Framework, for autonomous quality engineering on enterprise banking platforms. The framework binds a context engine (regulatory ontology, system and data topology, historical defect corpus, and requirement store) to six specialised agents (Context Manager, Test Planning, Test Generation, Test Execution and Self-Healing, Defect Triage and RCA, Compliance and Risk Scoring), coordinated over the Model Context Protocol and constrained by pre-execution and human-in-the-loop governance. The framework is grounded in an eleven-year, multi-affiliate-bank core-banking transformation and in an Agentic AI Test Framework serving over two thousand test cases across an integrated Salesforce, AWS, and SAP estate. A reference implementation of the Context Manager Agent and the regulatory rule-to-flow registry, together with practitioner-observed directional pre- and post-adoption ranges on defect leakage, automation coverage, cycle time, alert false-positive rate, and explainability compliance, illustrate operational behaviour. Values are illustrative and directional, not statistically controlled measurements; §8 states the specific validity threats a wider study would need to address. Findings are analytically, not statistically, generalisable.