InsureWise

Inspiration

Our team has all struggled with insurance: endless forms, confusing coverage, and no real clarity until it’s too late. Comparison sites just show prices, but don’t explain what you’re actually getting. We wanted to build a platform that truly understands your needs, explains coverage in plain language, and makes comparing policies simple and transparent.

What it does

InsureWise is an AI-powered insurance comparison and recommendation platform. Users chat with our AI to share their needs and priorities. The platform uses:

Conversational Onboarding: Captures your profile in a natural, chat-based flow.
AI-Powered Search: Fetches and scores policies from multiple insurers.
Priority-Weighted Ranking: Lets you adjust what matters most—price, coverage, or reputation—and instantly re-rank results.
AI Policy Explainer: Breaks down what’s covered, what’s missing, and highlights any gaps in plain English.
Premium Optimizer: Offers actionable tips to help you save on your policy.
Profile Management: Lets you update your details and re-run the optimizer anytime.

Clear, actionable recommendations to help you confidently choose the right insurance.

How we built it

InsureWise is built as a pnpm monorepo with shared types and a single OpenAPI spec, ensuring type safety and consistency across the stack.

Frontend: React, Vite, Tailwind CSS, shadcn/ui, Zustand for state management
Backend: Express 5 (Node.js) with PostgreSQL and Drizzle ORM
AI Chat: GPT-OSS 120B (OpenAI-compatible, HuggingFace endpoint)
RAG/Knowledge Assistant: Moorcheh AI (Python SDK) for semantic search and factual answers
Type Safety: OpenAPI → Orval codegen → Zod schemas and React Query hooks

Challenges we ran into

Reliable data extraction: Getting the AI to consistently extract structured profile data from free-form chat required careful prompt design and robust parsing.
Instant re-ranking: Achieving real-time, client-side re-ranking of policies based on user-set priorities meant pre-computing and normalizing scores server-side.
Monorepo setup: Integrating pnpm workspaces, Drizzle, Zod, and Vite with correct type sharing and path resolution took significant effort.

Accomplishments that we're proud of

A seamless, end-to-end flow from onboarding to personalized recommendations
AI-powered gap analysis that clearly shows users what’s missing in each policy
Clean, maintainable architecture with shared types and codegen

What we learned

Strategically combining fast, free AI models for chat with more advanced models for critical explanations leads to a better user experience and efficient resource use.