Cyber Catastrophe Bond Portfolio Modeling with CrewAI and Groq Cloud
Rising systemic cyber threats require multi-agent collaboration to stress-test catastrophe bonds. You will build a collaborative team of AI agents using CrewAI and Groq Cloud fast inference to model malware cascade scenarios across insurance portfolios. The solution must process 50 stress scenarios per minute with high agreement score (>0.90) across agent roles.
What you are building
Build a multi-agent cyber ILS stress-testing swarm using CrewAI orchestrated over high-throughput Groq Cloud inference.
How work is evaluated
Evaluates throughput of portfolio scenario processing and consensus quality among agents.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
high_consensus
Checks if multi-agent consensus_score exceeds 0.90.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
scenarios_per_minute
Number of simulated scenarios processed per minute • target: 50 • range: 0-100
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Orchestrate multi-agent roles (Threat Modeler, Actuary, Capital Manager) using CrewAI.
Configure Groq Cloud ultra-fast inference backends for low-latency agent iterations.
Simulate contagious malware propagate rates and compute probable maximum loss (PML).
Resources
Synthetic portfolio dataset comprising 1,000 commercial cyber risk policies with detailed cloud dependency mappings.
Challenge setup
Evaluates throughput of portfolio scenario processing and consensus quality among agents.
Challenge input
JSON object with scenario_name, affected_cloud_region, market_share_percentage
CrewAI
Assigned framework for role-playing multi-agent systems.
Groq Cloud
Assigned high-throughput LLM runtime platform.
Evaluated output
JSON object with total_estimated_pml, bond_principal_impairment_percent, consensus_score
- Checks if multi-agent consensus_score exceeds 0.90.
- Number of simulated scenarios processed per minute • target: 50 • range: 0-100
- Benchmark: Multi-Agent Consensus Benchmark
- Scenarios Per Minute target: 50
- 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
- CrewAI · Optional
- Llama 3.3 70B · Optional
- Groq Cloud · Required
Evaluation contract
- high_consensus · Weight 1
- scenarios_per_minute · Weight 1
Metrics
- scenarios_per_minute · Target 50
Benchmarks
- Multi-Agent Consensus Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with scenario_name, affected_cloud_region, market_share_percentage
Output: JSON object with total_estimated_pml, bond_principal_impairment_percent, consensus_score
Execution harness: Python
Evaluation tools
- CrewAI · Multi-Agent Framework
- Groq Cloud · Inference Acceleration Engine
Recipe freeze marker
No recipe freeze time is recorded.