Skip to main content
Loading navigation
Back to challenges

UPI Fraud Risk Triage Engine with Pydantic AI & Claude 4.1 Opus

Retail investment and fintech payment networks require sub-second fraud detection across high-volume UPI transactions. Build a real-time merchant risk engine using Pydantic AI and Claude 4.1 Opus to evaluate transaction metadata and flag synthetic fraud with 98% precision.

Business OperationsHosted by Vera
Sign in free to start

What you are building

Develop a type-safe UPI transaction fraud evaluation agent using Pydantic AI and Claude 4.1 Opus with Pydantic schema validation.

How work is evaluated

Evaluation

Evaluates accuracy of UPI fraud flagging and false positive rates under production load constraints.

Datasets

Public datasets, plus the private datasets you have access to.

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

pydantic_schema_strictness

Validates that all agent outputs comply exactly with output Pydantic model

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

fraud_detection_precision

Precision of blocked transactions matching gold label fraud cases • target: 0.98 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Master type-safe agent development using Pydantic AI dependencies and result validation

  • Leverage Claude 4.1 Opus for complex financial anomaly reasoning

  • Integrate device risk signals and merchant velocity metrics

  • Enforce strict SLA bounds (< 150ms reasoning overhead) via stream parsing

Resources

Dataset notes

UPI transaction telemetry stream with simulated mule VPAs, device hopping patterns, and merchant velocity metrics.

Challenge setup

Evaluates accuracy of UPI fraud flagging and false positive rates under production load constraints.

Challenge input

JSON transaction metadata with device ID, amount, and vpa_history

Pydantic AI

Type-safe agent framework for Python built on Pydantic.

Claude 4.1 Opus

High-reasoning LLM for complex anomaly and risk evaluation.

Evaluated output

JSON containing risk_score, action (ALLOW|BLOCK|FLAG), and reason

Evaluation checks
  • Validates that all agent outputs comply exactly with output Pydantic model
  • Precision of blocked transactions matching gold label fraud cases • target: 0.98 • range: 0-1
Configured evaluation evidence
  • Benchmark: UPI Financial Fraud Triage Benchmark
  • Fraud Detection Precision target: 0.98
  • 1 public reference case
Configured runtime
  • 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

Action Space
  • Pydantic AI · Required
  • BoTorch · Optional
Policy Serving
  • Claude 4.1 Opus · Optional
Orchestration
  • Pydantic AI · Required

Evaluation contract

  • pydantic_schema_strictness · Weight 1
  • fraud_detection_precision · Weight 1

Metrics

  • fraud_detection_precision · Target 0.98

Benchmarks

  • UPI Financial Fraud Triage Benchmark
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON transaction metadata with device ID, amount, and vpa_history

    Output: JSON containing risk_score, action (ALLOW|BLOCK|FLAG), and reason

    Execution harness: Python

Evaluation tools

  • Pydantic AI · Agent Orchestration Framework
  • Claude 4.1 Opus · Models

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about UPI Fraud Risk Triage Engine with Pydantic AI & Claude 4.1 Opus