LlamaIndex Advanced RAG Cyber Acquisition Due Diligence Engine
Munich Re's $575M acquisition of At-Bay highlights M&A expansion in cyber insurance. Build a LlamaIndex-powered RAG and agent workflow to ingest target cyber MGA underwriting portfolios, evaluate cyber catastrophe aggregation risks, and output automated due diligence audit scoring.
What you are building
Construct an enterprise LlamaIndex RAG application to perform cyber catastrophe due diligence on MGA target portfolios.
How work is evaluated
Evaluates retrieval accuracy and due diligence risk score computation across M&A target portfolios.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
bound_check
Ensures exposure percentage is between 0.0 and 1.0
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
retrieval_hit_rate
Recall of relevant risk factors from corpus • target: 0.88 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Implement LlamaIndex VectorStoreIndex and Document summary indices for target MGA portfolios
Build custom QueryEngineTools for cyber attack scenario stress testing
Deploy sub-question query engines to synthesize systemic cyber loss exposures
Calculate target company risk-adjusted valuation impact factors
Resources
Synthetic cyber insurance policy documentation and portfolio risk profiling reports.
Challenge setup
Evaluates retrieval accuracy and due diligence risk score computation across M&A target portfolios.
Challenge input
JSON object with target_mga_name and portfolio_summary document link.
LlamaIndex
Enterprise vector search and hierarchical document indices
Evaluated output
JSON containing systemic_cloud_exposure_pct, max_single_event_loss, and due_diligence_recommendation.
- Ensures exposure percentage is between 0.0 and 1.0
- Recall of relevant risk factors from corpus • target: 0.88 • range: 0-1
- Benchmark: CyberDueDateBench
- Retrieval Hit Rate target: 88%
- 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
- Exa · Optional
- Command R+ · Optional
Evaluation contract
- bound_check · Weight 1
- retrieval_hit_rate · Weight 1
Metrics
- retrieval_hit_rate · Target 0.88
Benchmarks
- CyberDueDateBench
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with target_mga_name and portfolio_summary document link.
Output: JSON containing systemic_cloud_exposure_pct, max_single_event_loss, and due_diligence_recommendation.
Execution harness: Python
Evaluation tools
- LlamaIndex · RAG Framework
Recipe freeze marker
No recipe freeze time is recorded.