Skip to main content
Loading navigation
Back to challenges

LlamaIndex Fraud Auditing RAG for Windstorm Claims

Insurers face systemic claims fraud following major hurricane events due to inconsistent damage documentation across post-event adjustment reports. Build a LlamaIndex RAG agent powered by Claude 4 Opus to analyze claim logs and flag fraudulent windstorm claims with 90% precision.

Machine LearningHosted by Vera
Sign in free to start

What you are building

Process claims notes and geospatial hurricane inspection logs using LlamaIndex vector indices and Claude 4 Opus for automated anomaly scoring.

How work is evaluated

Evaluation

Evaluates the RAG pipeline on fraud detection precision, citation accuracy, and retrieval latency.

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

schema_validation

Validates that the output JSON strictly matches required evaluation schema

Binary check

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

Dimension 2

precision

Precision of fraud detection on test dataset • target: 0.9 • 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

  • Implement a LlamaIndex knowledge graph and vector index over unstructured hurricane claims data

  • Leverage Claude 4 Opus reasoning capabilities to detect discrepancy patterns between field photos and adjuster text notes

  • Build a automated scoring pipeline that outputs auditable fraud likelihood ratings

Resources

Dataset notes

Synthetic dataset of 500 hurricane claim records containing adjuster descriptions, contractor invoices, and ground truth fraud tags.

Challenge setup

Evaluates the RAG pipeline on fraud detection precision, citation accuracy, and retrieval latency.

Challenge input

JSON containing claim text and field audit metadata

LlamaIndex

Used for vector indexing and structured query engines over claim documents

Claude 4 Opus

Provides high reasoning accuracy for document cross-referencing

Evaluated output

JSON with fraud_score (0.0-1.0), risk_level, and evidence_spans

Evaluation checks
  • Validates that the output JSON strictly matches required evaluation schema
  • Precision of fraud detection on test dataset • target: 0.9 • range: 0-1
Configured evaluation evidence
  • Benchmark: RAG-Insurance-Audit-Bench
  • Precision target: 0.9
  • 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
  • Llama Index · Required
  • Aim · Optional
Policy Serving
  • Claude 4 Opus · Optional

Evaluation contract

  • schema_validation · Weight 1
  • precision · Weight 1

Metrics

  • precision · Target 0.9

Benchmarks

  • RAG-Insurance-Audit-Bench
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing claim text and field audit metadata

    Output: JSON with fraud_score (0.0-1.0), risk_level, and evidence_spans

    Execution harness: Python

Evaluation tools

  • LlamaIndex · Orchestration Framework
  • Claude 4 Opus · LLM Engine

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about LlamaIndex Fraud Auditing RAG for Windstorm Claims