Build an Insurance Fraud Detection Knowledge Graph Agent with LlamaIndex
Cross-policy insurance fraud claims involve complex networks of claimants, repair shops, and medical providers. Develop a LlamaIndex Property Graph RAG agent that identifies fraudulent claim rings and outputs graph-based evidence packages for adjusters.
What you are building
Implement a LlamaIndex Property Graph RAG pipeline to uncover organized insurance claim fraud rings across linked entities.
How work is evaluated
Evaluates ring detection precision on collusive insurance claim graph datasets.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
Fraud Ring Threshold Test
Entities sharing >10 suspicious claims must trigger ring_detected == True.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Graph Ring Detection Recall
Percentage of true fraud rings identified. • target: 0.97 • range: 0.92-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Construct a LlamaIndex Property Graph Index from unstructured insurance loss claims.
Query multi-hop relationships between claimants, body shops, and medical clinics.
Detect collusive fraud rings with graph neural network heuristics and LLM synthesis.
Export structured evidence graphs for insurance fraud investigation teams.
Resources
Synthetic property graph dataset of connected motor and health insurance claims.
Challenge setup
Evaluates ring detection precision on collusive insurance claim graph datasets.
Challenge input
JSON containing claim_id, claimant_name, repair_shop_id, doctor_id, accident_location
LlamaIndex
Manages Property Graph construction and multi-hop graph retrieval.
Evaluated output
JSON with ring_detected (boolean), connected_claim_ids, evidence_graph_nodes
- Entities sharing >10 suspicious claims must trigger ring_detected == True.
- Percentage of true fraud rings identified. • target: 0.97 • range: 0.92-1
- Benchmark: InsuranceFraudGraphBench
- Graph Ring Detection Recall target: 0.97
- 1 public reference case
- 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
- Llama Index · Required
- LlamaIndex · Optional
- Llama 3.2 · Optional
Evaluation contract
- Fraud Ring Threshold Test · Weight 1
- Graph Ring Detection Recall · Weight 1
Metrics
- Graph Ring Detection Recall · Target 0.97
Benchmarks
- InsuranceFraudGraphBench
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing claim_id, claimant_name, repair_shop_id, doctor_id, accident_location
Output: JSON with ring_detected (boolean), connected_claim_ids, evidence_graph_nodes
Execution harness: Python
Evaluation tools
- LlamaIndex · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.