Build Multi-Agent Loan Risk Assessment Crew using CrewAI
Prudential's target $3 billion home improvement portfolio requires diverse financial expertise: property valuation analysts, credit risk evaluators, contractor quote auditors, and portfolio stress testers. Managing this requires a collaborative multi-agent workforce. In this challenge, developers use CrewAI to orchestrate a specialized crew of AI agents (Underwriting Lead, Property Appraiser, Contractor Verification Specialist, Compliance Auditor). The crew collaborates dynamically to evaluate complex home improvement loan applications, detecting potential contractor fraud and verifying post-renovation loan collateral security.
What you are building
Orchestrate a multi-agent workforce with CrewAI to audit home improvement loan portfolios and verify renovation estimates.
How work is evaluated
Evaluates multi-agent agreement, detection of inflated contractor estimates, and accuracy of final portfolio risk assignment.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
Contractor Overquote Detection
Verifies crew successfully identifies inflated kitchen remodeling quote
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Risk Flag Recall
Accuracy of detected risk indicators during multi-agent audit • target: 0.92 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Define specialized agent roles, goals, and backstories using CrewAI
Implement CrewAI tasks and sequential execution flows for multi-agent collaboration
Build custom contractor quote verification tools to detect inflated renovation estimates
Consolidate agent assessments into an institutional loan investment memo
Resources
Dataset of 50 home improvement loan packages with line-item contractor bids and property appraisals.
Challenge setup
Evaluates multi-agent agreement, detection of inflated contractor estimates, and accuracy of final portfolio risk assignment.
Challenge input
JSON with applicant_credit_score, property_address, contractor_quote, scope_of_work
CrewAI
Required framework for multi-agent role play.
Evaluated output
JSON with final_recommendation ('APPROVE', 'DENY', 'ADJUST_AMOUNT'), approved_loan_amount, detected_risk_flags
- Verifies crew successfully identifies inflated kitchen remodeling quote
- Accuracy of detected risk indicators during multi-agent audit • target: 0.92 • range: 0-1
- Benchmark: MultiAgentUnderwriteBench
- Risk Flag Recall target: 0.92
- 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
- Zed · Optional
Evaluation contract
- Contractor Overquote Detection · Weight 1
- Risk Flag Recall · Weight 1
Metrics
- Risk Flag Recall · Target 0.92
Benchmarks
- MultiAgentUnderwriteBench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with applicant_credit_score, property_address, contractor_quote, scope_of_work
Output: JSON with final_recommendation ('APPROVE', 'DENY', 'ADJUST_AMOUNT'), approved_loan_amount, detected_risk_flags
Execution harness: Python
Evaluation tools
- CrewAI · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.