Agentic Transaction Fraud & Sanctions Triage Engine with Pydantic AI and Azure AI
Autonomous AI trading agents and high-frequency settlement channels require real-time risk controls capable of making microsecond compliance decisions. Using Pydantic AI's type-safe agent framework deployed alongside Azure AI Services, developers can build deterministic, inspectable payment screening agents that detect complex sanctions evasions and high-risk trading velocity pattern spikes.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a type-safe agentic transaction triage system using Pydantic AI and Azure AI to validate high-velocity trading requests against global sanctions list databases.
How work is evaluated
Tests Pydantic AI model output validation and detection accuracy for sanctioned entities attempting trade routing under modified names.
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.
pydantic_validation_check
Output must deserialize without Pydantic ValidationError.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
sanctions_hit_recall
Recall metric for detecting known sanctioned entities. • target: 1 • range: 0.9-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Construct strict Pydantic AI schemas for financial transaction validation
Integrate Azure AI Search and Azure OpenAI for fuzzy sanctions list matching
Implement dependency injection in Pydantic AI for dynamic database rule lookup
Deploy automated fraud confidence scoring with human-in-the-loop fallback hooks
Reference links and supporting material
Dataset of 2,000 synthetic algorithmic trading orders containing obfuscated names, variations in transliteration, and embedded sanctioned entities.
Challenge setup
Tests Pydantic AI model output validation and detection accuracy for sanctioned entities attempting trade routing under modified names.
Challenge input
JSON transaction payload with sender_name, ultimate_beneficiary, asset, and order_value.
Pydantic AI
Ensures deterministic schema validation for trade approval outputs.
Azure AI Services
Provides scalable vector indexing and model execution runtime.
Evaluated output
Validated Pydantic JSON with status (APPROVED|FLAGGED|BLOCKED), match_confidence, and regulatory_rule_id.
- Output must deserialize without Pydantic ValidationError.
- Recall metric for detecting known sanctioned entities. • target: 1 • range: 0.9-1
- Benchmark: Pydantic AI Compliance Benchmark
- Sanctions Hit Recall 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
- Azure · Required
- Azure OpenAI Service · Optional
Evaluation contract
- pydantic_validation_check · Weight 1
- sanctions_hit_recall · Weight 1
Metrics
- sanctions_hit_recall · Target 1
Benchmarks
- Pydantic AI Compliance Benchmark
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON transaction payload with sender_name, ultimate_beneficiary, asset, and order_value.
Output: Validated Pydantic JSON with status (APPROVED|FLAGGED|BLOCKED), match_confidence, and regulatory_rule_id.
Execution harness: Python
Evaluation tools
- Pydantic AI · Type-Safe Agent Framework
- Azure AI Services · Inference and Hybrid Search
Recipe freeze marker
No recipe freeze time is recorded.