Skip to main content
Loading navigation
Back to challenges

CrewAI & Triton Inference Server Soft Market Commercial Underwriting Orchestrator

During soft market conditions, commercial property insurers face competitive premium pressure demanding hyper-precise risk pricing. Build a CrewAI agent team backed by high-throughput ML models served via Triton Inference Server to optimize commercial policy pricing and underwriting guidelines.

Machine LearningHosted by Vera
Sign in free to start

What you are building

Orchestrate a CrewAI multi-agent team with Triton Inference Server models to optimize commercial insurance pricing in soft markets.

How work is evaluated

Evaluation

Evaluates commercial quote generation latency, target loss ratio preservation, and market competitive pricing.

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

loss_ratio_guardrail

Ensures estimated loss ratio is below target maximum

Binary check

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

Dimension 2

triton_latency_ms

Inference roundtrip time • target: 50 • range: 0-200

Binary check

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

Learning goals

  • Design CrewAI agents with dedicated underwriting roles (Market Analyst, Risk Evaluator, Pricing Agent)

  • Serve ONNX/PyTorch pricing models using NVIDIA Triton Inference Server

  • Implement gRPC tool calls connecting CrewAI agents to Triton model endpoints

  • Optimize rate recommendations under market soft-rate pressure constraints

Resources

Dataset notes

Commercial lines property exposure dataset with historical underwriting loss outcomes.

Challenge setup

Evaluates commercial quote generation latency, target loss ratio preservation, and market competitive pricing.

Challenge input

JSON object describing commercial account exposure and competitor benchmark rate.

CrewAI

Role-playing multi-agent collaboration

Triton Inference Server

High-throughput performant model serving

Evaluated output

JSON containing proposed premium rate, model loss ratio estimate, and underwriting rationale.

Evaluation checks
  • Ensures estimated loss ratio is below target maximum
  • Inference roundtrip time • target: 50 • range: 0-200
Configured evaluation evidence
  • Benchmark: CommercialPricingBench
  • Triton Latency Ms target: 50
  • 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
  • Triton Inference Server · Required
  • Triton Inference Server · Optional
  • CrewAI · Optional

Evaluation contract

  • loss_ratio_guardrail · Weight 1
  • triton_latency_ms · Weight 1

Metrics

  • triton_latency_ms · Target 50

Benchmarks

  • CommercialPricingBench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON object describing commercial account exposure and competitor benchmark rate.

    Output: JSON containing proposed premium rate, model loss ratio estimate, and underwriting rationale.

    Execution harness: Python

Evaluation tools

  • CrewAI · Agent Framework
  • Triton Inference Server · Model Runtime

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about CrewAI & Triton Inference Server Soft Market Commercial Underwriting Orchestrator