International Journal of Scientific Research and Engineering Development

The adjudication of credit risk in the mortgage industry is a knowledge-intensive process constrained by the dual imperatives of high efficiency and unwavering regulatory compliance. Existing AI solutions, often monolithic "black box" models, excel at prediction but fail to meet the critical demands for explainability, auditability, and human-in-the-loop collaboration. This paper introduces the LoanApprovalEngine, a comprehensive, Python-based Hybrid Intelligence framework that re-conceptualizes underwriting as a collaborative task between a human expert and a multi-agent AI system. Our architecture is composed of three specialized, communicating agents:
(1) A Prognostic Agent (MLRiskScorer, InterestRateModel), built on an inherently transparent, rules-based model that generates risk scores along with a causal chain of explanatory factors.
(2) A Narrative Agent (_generate_ai_credit_memo), a generative LLM that synthesizes the Prognostic Agent's structured output into formal credit appraisal memos.
(3) An Interrogative Agent (voice-enabled chatbot), which facilitates a human-in-the-loop hermeneutic process, allowing underwriters to conduct dynamic, multilingual analysis on filtered applicant cohorts. We provide a detailed analysis of how the integration of AI tools—including sentence-transformers for data ingestion, pyttsx3 and langdetect for multilingual voice, and LLMs for reasoning and generation— transforms the underwriting workflow. We demonstrate that this framework drastically reduces processing time for bulk assessments and enables a depth of interactive analysis unattainable through traditional methods.

Nagraj Gangadhar Halburge, Navika Joshi, Khizar Khan, Sudhanva Kulkarni, "A Synergistic Multi-Agent AI Framework for Cognitive Augmentation in Home Loan Credit Risk Assessment" International Journal of Scientific Research and Engineering Development, V8(6): Page(2675-2687) Nov-Dec 2025. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.