Skip to main content
Back to challenges
Challenge

AutoGen Multi-Agent Underwriting System with Cerebras

Modern insurance carriers require high-throughput multi-agent systems to perform immediate policy risk assessments across thousands of applications. Construct an AutoGen conversational multi-agent underwriting framework using Cerebras Inference to process 500 risk profiles per second with under 100ms latency.

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.

Coordinate AutoGen specialized agents using Cerebras fast inference endpoints to evaluate personal property coverage guidelines.

Delivery guide

How work is evaluated

Evaluation

Measures multi-agent throughput and risk assessment accuracy against benchmark underwriting decisions.

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

latency_check

Ensures average decision latency is below 100ms per profile

Binary check

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

Dimension 2

throughput_qps

Queries processed per second • target: 500 • range: 100-1000

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

  • Design an AutoGen agent conversation graph for parallel property risk classification

  • Integrate Cerebras high-speed Llama-3 inference endpoints for ultra-low latency decision making

  • Validate policy compliance and calculate custom premium rate adjustments

Resources and assets

Reference links and supporting material

Dataset notes

Dataset of 2000 property underwriting profiles including roof age, flood zone ratings, and claim history.

Challenge setup

Measures multi-agent throughput and risk assessment accuracy against benchmark underwriting decisions.

Challenge input

List of property risk profiles

AutoGen

Provides multi-agent conversation patterns and role segregation

Cerebras Inference

Enables ultra-fast token generation necessary for low-latency multi-agent loops

Evaluated output

JSON containing pass/fail rate, average latency, and decision details

Evaluation checks
  • Ensures average decision latency is below 100ms per profile
  • Queries processed per second • target: 500 • range: 100-1000
Configured evaluation evidence
  • Benchmark: MultiAgent-Speed-Bench
  • Throughput Qps target: 500
  • 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
  • AutoGen · Required
  • Llama 3.1 405B · Optional
Policy Serving
  • Cerebras Inference · Optional

Evaluation contract

  • latency_check · Weight 1
  • throughput_qps · Weight 1

Metrics

  • throughput_qps · Target 500

Benchmarks

  • MultiAgent-Speed-Bench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: List of property risk profiles

    Output: JSON containing pass/fail rate, average latency, and decision details

    Execution harness: Python

Evaluation tools

  • AutoGen · Agent Framework
  • Cerebras Inference · Inference Acceleration Engine

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about AutoGen Multi-Agent Underwriting System with Cerebras