The purpose, in one line: auto quote generation and agentic underwriting. A customer finishes a property quote in minutes; behind the desk, AI agents assemble the risk picture and draft the decision — and a human underwriter owns it.
This is a proof of concept for a property-insurance startup idea — not a production system — built on years of group-benefits insurance platform experience (rules engines, underwriting workflows). It showcases how agentic systems can be built around the full insurance pipeline: quote → underwriting → servicing → claims. Mock data throughout; every flow below is clickable.
QuoteCraft (the Zen Algorithms property-insurance platform) runs the full property/fire lifecycle — one engine, three role-based workspaces over a shared data spine (one event log all workspaces read from). A product is metadata + rules (line of business → policy form → provisions → options); compliance and risk/underwriting rules evaluate live; a climate cat-model (CAT = catastrophe) scores each address and an AI agent (RAG — Retrieval-Augmented Generation — knowledge system) generalizes underwriter decisions across a geography. Pick a role to begin.
Quote-to-submit journey: sign in, onboard the insured & property, choose a line, design the policy, and submit it for underwriting.
How business arrives: customers buy direct, producers initiate on a customer's behalf, and mortgage servicers track evidence of insurance — with staged identity gates on paid data pulls and rules for when channels collide on one property.
Work the referral queue (submissions automated underwriting kicked out for human review): open a submission, read its cat-model (catastrophe-model) score & flagged rules, approve/refer/decline, then price & bind — and train the AI agent on the geography.
Service in-force policies and handle losses: review the bound case & endorsements, take FNOL (First Notice of Loss), and let claims tag risk profiles back into the RAG (Retrieval-Augmented Generation) knowledge system.
The live console where the agents work: submissions triaged by an agent loop, risk evidence gathered by tools, a multi-model council drafting the underwriting decision — and a human approve/refer/decline gate on every binding action. Live model calls are enabled per demo session; the full workflow, reasoning traces, and audit log are always visible.
How it's built: the five-flow event-sourced data spine, climate / VLM (Vision-Language Model) / RAG (Retrieval-Augmented Generation) integrations, and a GCP-vs-AWS backend with deploy strategy — every decision annotated, every term clickable.
The simplified view: the same platform explained one flow at a time — what each role touches, in plain language, without the full systems diagram.