Technology
Every policy we write is backed by a proprietary AI catastrophe model that sees risk at a resolution legacy vendors can't match. Spatial intelligence, physics-based modeling, and end-to-end pricing — purpose-built for catastrophe insurance.
PigeonNet
Legacy cat models treat every building as an island. Ours understands how catastrophes actually behave — spatially connected, physics-driven, and always uncertain.
Our model understands that properties don't exist in isolation. Risk spreads spatially — fire jumps between buildings, wind cascades through corridors, water follows terrain. We capture these connections at multiple scales, from individual structures to entire regions.
We don't rely on historical averages or ZIP-code-level statistics. Our catastrophe engine is trained on real physics simulations — modeling how fire actually spreads, how wind pressure loads structures, and how water routes through terrain. Orders of magnitude faster than traditional methods.
Every pricing decision is informed by the full portfolio context. Deductibles, limits, coinsurance, and reinsurance structures are all mathematically connected to the underlying risk. We can quantify exactly how policy terms affect expected loss.
Every quote includes calibrated confidence bounds — not just a point estimate. We know what we know, and more importantly, we know what we don't. Actuaries and regulators appreciate the difference.
See What Our Models See
When catastrophe strikes, Pigeon's AI has already identified which properties can withstand the risk. Drag the slider to see how our models highlight insurable properties — even after a wildfire.
AI Risk AssessmentAI Architecture
Speed is table stakes. We built an AI infrastructure designed for the regulated, high-stakes world of insurance — where every decision needs to be explainable, auditable, and safe.
Every AI call passes through a layered security stack — input validation, output scanning, cost controls, and full audit logging. Security is architectural, not an afterthought.
Purpose-built AI skill templates for every domain task — appetite evaluation, quote generation, compliance documents, claim triage. Structured, repeatable, and auditable.
Long-running workflows that survive failures gracefully. Each step is individually retriable. Human approval gates can pause a workflow for weeks without losing state.
An AI copilot with real-time access to submission data, policy details, address geocoding, and a help knowledge base. Answers questions in context, not from a script.
A hybrid search system that combines semantic understanding with keyword precision. Finds the right answer even when the question doesn't use the exact right words.
We stand on the shoulders of billions in foundation model research — leveraging state-of-the-art models for geographic understanding, satellite imagery analysis, and language intelligence.
Under the Hood