CrewAI Multi-Agent GAIP Insurance Risk Underwriter
Orchestrate a team of CrewAI agents with distinct roles (risk assessor, policy analyst, auditor) to process commercial insurance claims and calculate risk scores. Achieve zero policy hallucination errors and automated audit logging.
What you are building
Design a collaborative multi-agent system using CrewAI with defined roles and delegated tasks for underwriting regional commercial insurance.
How work is evaluated
Evaluates multi-agent CrewAI task completion accuracy and calculated payout precision.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
Math Calculation Correctness
Verifies payout calculation accurately factors in deductibles
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
payout_accuracy
Accuracy of payout recommendation • target: 1 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Configure CrewAI agents with distinct backstories, goals, and role capabilities
Implement sequential and hierarchical agent delegation processes
Enforce compliance and coverage boundaries for regional commercial insurance
Generate structured claim settlement reports with full rationale
Resources
Synthetic commercial property and casualty claim files tailored to Asian insurance markets.
Challenge setup
Evaluates multi-agent CrewAI task completion accuracy and calculated payout precision.
Challenge input
JSON containing policy terms, claim loss details, and surveyor report
CrewAI
Framework for orchestrating role-playing collaborative agents
Evaluated output
JSON containing approved_amount, deduction_reasons, and crew_task_status
- Verifies payout calculation accurately factors in deductibles
- Accuracy of payout recommendation • target: 1 • range: 0-1
- Benchmark: InsurBench
- Payout Accuracy target: 1
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- CrewAI · Required
- crewAI · Optional
- Aim · Optional
Evaluation contract
- Math Calculation Correctness · Weight 1
- payout_accuracy · Weight 1
Metrics
- payout_accuracy · Target 1
Benchmarks
- InsurBench
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing policy terms, claim loss details, and surveyor report
Output: JSON containing approved_amount, deduction_reasons, and crew_task_status
Execution harness: Python
Evaluation tools
- CrewAI · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.