Skip to main content
Back to challenges
Challenge

Claude Agents SDK & Featherless AI Policy Audit Engine for LLM Underwriting

Insurers like Allstate are deploying proprietary LLMs like ALLIE for carrier workflows. Build an audited policy underwriting QA pipeline using the Claude Agents SDK hosted with Featherless AI inference endpoints to verify model responses against statutory compliance benchmarks with 95% audit coverage.

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.

Develop an audited policy document QA agent using Claude Agents SDK and Featherless AI to validate insurance assistant responses.

Delivery guide

How work is evaluated

Evaluation

Tests policy advice accuracy and citation validity across underwriting scenarios.

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

citation_exactness

Ensures cited section exists in input policy text

Binary check

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

Dimension 2

accuracy_score

Proportion of correct policy answers • target: 0.95 • 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

What you should walk away with

  • Configure Claude Agents SDK with custom tools for statutory policy compliance parsing

  • Route open-weights and custom inference models through Featherless AI endpoints

  • Implement automated policy citation and hallucination detection tools

  • Measure coverage and accuracy against standardized carrier Q&A benchmarks

Resources and assets

Reference links and supporting material

Dataset notes

Dataset of standard ISO commercial property and casualty policy forms with ground-truth QA annotations.

Challenge setup

Tests policy advice accuracy and citation validity across underwriting scenarios.

Challenge input

JSON object with customer query and policy document context.

Claude Agents SDK

Agent logic and function calling

Featherless AI

Serverless open model inference execution

Evaluated output

JSON object with answer, exact clause citations, and compliance audit pass status.

Evaluation checks
  • Ensures cited section exists in input policy text
  • Proportion of correct policy answers • target: 0.95 • range: 0-1
Configured evaluation evidence
  • Benchmark: PolicyQA-Bench
  • Accuracy Score target: 0.95
  • 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

Environment
  • Replit Agent · Optional
Action Space
  • Featherless AI · Required
  • Haize Labs · Optional
  • Replit Agent · Optional

Evaluation contract

  • citation_exactness · Weight 1
  • accuracy_score · Weight 1

Metrics

  • accuracy_score · Target 0.95

Benchmarks

  • PolicyQA-Bench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON object with customer query and policy document context.

    Output: JSON object with answer, exact clause citations, and compliance audit pass status.

    Execution harness: Python

Evaluation tools

  • Claude Agents SDK · Agent Framework
  • Featherless AI · Inference Host

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Claude Agents SDK & Featherless AI Policy Audit Engine for LLM Underwriting