Build Home Improvement Loan Portfolio Underwriter with Pydantic AI & Agno
Prudential's asset arm is targeting a $3 billion home improvement loan portfolio. Managing collateral risk, property equity estimations, and borrower debt-to-income limits at scale requires strict structured data validation and type-safe agent execution. In this challenge, developers will build a production-grade loan portfolio underwriting pipeline using Pydantic AI and Agno. The agent processes incoming loan packages, validates contractor proposals, calculates dynamic Loan-to-Value (LTV) post-renovation, and enforces compliance policies.
What you are building
The core problem, expected build, and operating context for this challenge.
Construct a type-safe loan underwriting agent using Pydantic AI and Agno to evaluate a $3B home improvement loan portfolio.
How work is evaluated
Evaluates loan application extraction accuracy, ARV/LTV validation, and schema compliance under strict type checking.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
CLTV Calculation
Ensures CLTV = (Existing Mortgage + Requested Loan) / ARV matches target within 0.01
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Schema Validation Compliance
Percentage of generated outputs strictly satisfying Pydantic schema • target: 1 • range: 1-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Design strict Pydantic data schemas for renovation project quotes and property valuations
Build a multi-step agent flow using Pydantic AI with automatic error recovery and dynamic schema validation
Integrate Agno autonomous agent infrastructure for scalable parallel portfolio processing
Implement post-renovation ARV (After Repair Value) risk scoring algorithms
Reference links and supporting material
Dataset containing 200 home improvement loan applications with contractor estimates and appraisal records.
Challenge setup
Evaluates loan application extraction accuracy, ARV/LTV validation, and schema compliance under strict type checking.
Challenge input
JSON with property_value, existing_mortgage, requested_loan, renovation_cost, estimated_value_add, monthly_income, existing_debts
Pydantic AI
Required framework for typed agent responses.
Agno
Required tool for agent orchestration.
Evaluated output
JSON with decision ('APPROVED', 'REJECTED', 'MANUAL_REVIEW'), cltv, arv, max_allowed_loan
- Ensures CLTV = (Existing Mortgage + Requested Loan) / ARV matches target within 0.01
- Percentage of generated outputs strictly satisfying Pydantic schema • target: 1 • range: 1-1
- Benchmark: UnderwriteBench 2025
- Schema Validation Compliance target: 1
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Agno · Required
- Pydantic AI · Optional
- LiveKit · Optional
- Agno · Required
- Pydantic AI · Optional
Evaluation contract
- CLTV Calculation · Weight 1
- Schema Validation Compliance · Weight 1
Metrics
- Schema Validation Compliance · Target 1
Benchmarks
- UnderwriteBench 2025
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with property_value, existing_mortgage, requested_loan, renovation_cost, estimated_value_add, monthly_income, existing_debts
Output: JSON with decision ('APPROVED', 'REJECTED', 'MANUAL_REVIEW'), cltv, arv, max_allowed_loan
Execution harness: Python
Evaluation tools
- Pydantic AI · Agent Framework
- Agno · Agent Runtime
Recipe freeze marker
No recipe freeze time is recorded.