GenAI Risk Intelligence Platform
A GenAI-assisted risk intelligence workflow that connects portfolio data and catastrophe-model outputs to natural-language querying, grounded summaries, and decision-support insights. The design emphasizes traceable evidence, integration with existing analytical workflows, and clear boundaries between generated explanations and underlying model results.
Natural language query
Interface
CAT model + exposure data
Grounding
Cited narrative insights
Output
Problem & use case
Risk analysts spent days manually compiling portfolio insights from disparate CAT model outputs, exposure files, and event response reports — slowing down underwriting and post-event decisioning.
Solution & engineering approach
A LangChain-orchestrated agent layer via Amazon Bedrock connected to internal exposure databases and CAT model outputs, producing instant narrative insights, executive summaries, and what-if scenario explanations.
Architecture & data flow
- 01
Ingest portfolio + CAT model outputs into structured store
- 02
Vectorize exposure metadata and historical event reports
- 03
LangChain agent routes user queries to SQL, vector, and tool nodes
- 04
LLM via Amazon Bedrock generates grounded narrative + recommendations
- 05
React dashboard renders insights, citations, and scenario deltas
Technology stack
Implementation status
Implemented in an enterprise risk analytics context — specific performance metrics not disclosed; description reflects system capability and design