Skip to main content
Loading navigation
Back to challenges

Global Reinsurance ILS Risk Model using Claude Agents SDK and Azure

Reinsurance treaties and Casualty ILS require complex risk aggregation and exposure profiling across global portfolios. Build an automated underwriting agent using Claude Agents SDK on Azure to evaluate counterparty exposure and catastrophe risk concentration. Achieve a 92% calibration accuracy score against historical loss benchmarks.

Machine LearningHosted by Vera
Sign in free to start

What you are building

Build an automated reinsurance ILS risk modeling agent with Claude Agents SDK deployed on Azure infrastructure for complex portfolio loss aggregation.

How work is evaluated

Evaluation

Evaluates risk aggregation and calibration accuracy on casualty ILS portfolio loss scenarios.

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

valid_risk_metrics

Expected Loss must be non-negative and within valid mathematical bounds.

Binary check

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

Dimension 2

calibration_score

Model risk calibration score against historical loss distributions • target: 0.92 • range: 0.7-1

Binary check

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

Learning goals

  • Build Claude Agents SDK structured workflow with extended thinking capabilities

  • Deploy agent pipeline on Azure AI infrastructure with secure data connections

  • Aggregate risk exposure for Casualty ILS bonds and reinsurance treaties

  • Measure risk calibration score against benchmark disaster loss scenarios

Resources

Dataset notes

Anonymized global reinsurance loss portfolio dataset containing 1,000 catastrophe and casualty treaty structures.

Challenge setup

Evaluates risk aggregation and calibration accuracy on casualty ILS portfolio loss scenarios.

Challenge input

JSON representing portfolio treaty limits, attachment points, and underlying perils

Claude Agents SDK

Provides deep reasoning and tool integration for complex financial treaty parsing.

Azure

Enterprise cloud hosting environment for secure agent model orchestration.

Evaluated output

JSON containing expected_loss_pct (float), tvar_99 (float), and calibration_score (float)

Evaluation checks
  • Expected Loss must be non-negative and within valid mathematical bounds.
  • Model risk calibration score against historical loss distributions • target: 0.92 • range: 0.7-1
Configured evaluation evidence
  • Benchmark: ReinsuranceRiskBench
  • Calibration Score target: 0.92
  • 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
  • Azure · Required
  • DeepSeek · Optional
  • Lepton AI · Optional

Evaluation contract

  • valid_risk_metrics · Weight 1
  • calibration_score · Weight 1

Metrics

  • calibration_score · Target 0.92

Benchmarks

  • ReinsuranceRiskBench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON representing portfolio treaty limits, attachment points, and underlying perils

    Output: JSON containing expected_loss_pct (float), tvar_99 (float), and calibration_score (float)

    Execution harness: Python

Evaluation tools

  • Claude Agents SDK · Agent Framework
  • Azure · AI Engineering Tooling

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Global Reinsurance ILS Risk Model using Claude Agents SDK and Azure