Build an Industrial Complex Financing Agent with Pydantic AI
Structured debt financing for industrial parks like New October ($12m+) requires precise covenant monitoring. Build a type-safe debt monitoring agent using Pydantic AI to parse loan agreements, track drawdown schedules, and enforce debt service coverage ratios.
What you are building
Utilize Pydantic AI type-safe agent models to analyze bank financing covenants and monitor industrial park drawdown compliance.
How work is evaluated
Evaluates loan covenant validation accuracy and drawdown approval enforcement.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
covenant_compliance_check
Ensure tranche is approved only when construction threshold is met
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
covenant_extraction_accuracy
Accuracy of extracted covenant variables • target: 0.99 • range: 0.9-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Define strict Pydantic schemas for bank credit facilities and repayment terms
Extract financial covenants (LTV, DSCR) from legal loan agreement text
Automate drawdown eligibility checks against project construction milestones
Output verified covenant compliance certificates for bank credit committees
Resources
Industrial development loan agreements, construction drawdown certificates, and ADCB financing covenant disclosures.
Challenge setup
Evaluates loan covenant validation accuracy and drawdown approval enforcement.
Challenge input
JSON milestone and covenant data
Pydantic AI
Type-safe agent framework ideal for strict financial output verification
Evaluated output
JSON drawdown approval
- Ensure tranche is approved only when construction threshold is met
- Accuracy of extracted covenant variables • target: 0.99 • range: 0.9-1
- Benchmark: FinCovenantBench
- Covenant Extraction Accuracy target: 0.99
- 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
- Pydantic AI · Required
- Coval · Optional
- Bito AI · Optional
- Pydantic AI · Required
Evaluation contract
- covenant_compliance_check · Weight 1
- covenant_extraction_accuracy · Weight 1
Metrics
- covenant_extraction_accuracy · Target 0.99
Benchmarks
- FinCovenantBench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON milestone and covenant data
Output: JSON drawdown approval
Execution harness: Python
Evaluation tools
- Pydantic AI · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.