The Solution: Leveraging Agentic AI for Digital Underwriting
To address these challenges, a multi-agent AI underwriting system was implemented to automate end-to-end financial and medical risk assessment. The platform uses specialized AI agents that collaborate to evaluate applicant data, enforce underwriting rules, detect fraud, and generate final underwriting decisions.
5-Agent AI Architecture:
Document Analysis Agent
Extracts and structures financial and medical data from bank statements, salary slips, ITR, Form 16A, PAN/Aadhaar documents, lab reports, MER, hematology reports, urine analysis, and ECG through Intelligent Document Processing (TruCap+). It converts unstructured records into standardized data for analysis.
Financial Risk Assessment Agent
Evaluates applicant financial eligibility using Human Life Value (HLV), Total Financial Sum At Risk (TFSAR), and Medical Sum At Risk (MSAR) calculations. It analyzes premium-to-income ratios, affordability, age-based income multiples, and verifies bank statements, salary slips, and financial documents through identity cross-matching.
Medical Risk Assessment Agent
Performs health evaluation through BMI calculation, hematology analysis (hemoglobin, RBC, WBC), blood pressure and lab value assessment, and age–sum assured-based medical grid matching. The agent generates health scores and flags abnormal conditions for consistent medical risk evaluation.
Fraud Detection Agent
Identifies suspicious submissions through diagnostic centre blacklist verification, geographic distance analysis between home address and diagnostic locations, and detection of identical test result patterns. Configurable thresholds classify fraud risk levels.
Decision Support Agent
Consolidates insights from all agents to categorize risk levels (low, moderate, high) and generate final underwriting decisions such as approved, declined, pending documents, or under review, along with clear recommendations and required actions.
Rule-Based Intelligence & Compliance:
The platform dynamically interprets financial and medical underwriting rules, including age based income multiples, MSAR-based medical requirements, and medical test grids. Using LLM-powered parsing, the system reads underwriting rules directly from uploaded DOCX documents, enabling dynamic rule updates, selective rule application based on applicant profile, and version control with rollback capability.
Technology Foundation:
The solution is powered by a robust enterprise technology stack built on a multi-agent AI architecture that enables coordinated decision-making across underwriting functions. Agent workflows are orchestrated using the CrewAI framework, while Azure OpenAI (GPT-4o) models drive deterministic rule parsing and risk evaluation. Seamless integration with TruCap+ enables automated data ingestion from documents. The system operates on a FastAPI backend with secure authentication, audit logging, batch processing capabilities, and configurable agent controls. This architecture ensures scalability, transparency, and compliance-ready deployment with a complete audit trail of decisions and rule updates.
