Skip to main content
Back to challenges
Challenge

TWIA Hurricane Claims Fraud Detector with OpenAI Agents SDK

Post-storm insurance fraud surges during catastrophic events like Hurricane Beryl, driving up litigation and carrier losses. Build a fraud detection workflow using OpenAI Agents SDK and GPT-5 Pro that ingests claims notes and building inspections to classify fraudulent damage patterns. Achieve 90% ROC-AUC on post-disaster claims evaluation.

Machine LearningHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Build an agentic claims fraud detection pipeline using the OpenAI Agents SDK and GPT-5 Pro function calling to inspect hurricane claims documentation.

Delivery guide

How work is evaluated

Evaluation

Evaluates claims fraud classification accuracy and output schema compliance.

Datasets

Challenge datasets

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

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

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

schema_validity

Output must match expected JSON schema.

Binary check

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

Dimension 2

roc_auc

Area under ROC curve for fraud detection • target: 0.9 • range: 0.5-1

Binary check

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

Learning goals

What you should walk away with

  • Implement multi-agent workflow routing using OpenAI Agents SDK

  • Integrate GPT-5 Pro function calling for claims note parsing

  • Extract structured fraud signals from unstructured catastrophe inspection reports

  • Evaluate classification accuracy using ROC-AUC on real-world post-storm datasets

Resources and assets

Reference links and supporting material

Dataset notes

Synthetic dataset of 500 post-hurricane property claims including adjuster text notes, photo descriptions, and contractor estimates.

Challenge setup

Evaluates claims fraud classification accuracy and output schema compliance.

Challenge input

JSON containing claim_id, damage_description, contractor_quote, and inspection_notes

OpenAI Agents SDK

Orchestrates stateful multi-agent communication and tool execution.

GPT-5 Pro

Provides high-reasoning accuracy for complex insurance contract and claim parsing.

Evaluated output

JSON with fraud_probability (float 0-1), flag_reason (string), and action (string)

Evaluation checks
  • Output must match expected JSON schema.
  • Area under ROC curve for fraud detection • target: 0.9 • range: 0.5-1
Configured evaluation evidence
  • Benchmark: ClaimsFraudBench-2026
  • Roc Auc target: 0.9
  • 1 public reference case
Configured runtime
  • 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

Action Space
  • OpenAI · Required
  • Aim · Optional
  • Seldon Core · Optional

Evaluation contract

  • schema_validity · Weight 1
  • roc_auc · Weight 1

Metrics

  • roc_auc · Target 0.9

Benchmarks

  • ClaimsFraudBench-2026
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing claim_id, damage_description, contractor_quote, and inspection_notes

    Output: JSON with fraud_probability (float 0-1), flag_reason (string), and action (string)

    Execution harness: Python

Evaluation tools

  • OpenAI Agents SDK · Agent Framework
  • GPT-5 Pro · LLM Engine

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about TWIA Hurricane Claims Fraud Detector with OpenAI Agents SDK