Insurance · Applied AI · Product
COVER — AI-Native Insurance Platform
Built an AI-native insurance platform that turns product information, images, and proof of purchase into personalized coverage options through automated assessment, underwriting, pricing, and policy generation.
- Role
- AI Underwriting System
- Date
- 2026
- Context
- AI underwriting system for second-hand goods
Technologies
Context
Context
Traditional item insurance can involve significant friction between identifying the asset, assessing eligibility, pricing risk, selecting coverage, and creating a policy. COVER explores what this workflow could look like if those steps were designed around modern artificial intelligence (AI) and programmable underwriting from the beginning. It is an AI-native insurance experience for individual physical items, particularly second-hand goods — a working product demo, not a licensed insurance carrier.
Problem
Problem
How can an individual physical item be turned into an insurable object from minimal user input?
Pricing and underwriting need to stay structured — a generative model should not invent a price or a coverage decision.
Messy multimodal inputs (a product URL, photos, a receipt) have to become a clear set of options a person can actually choose from.
Scope
System scope
Product input from a URL, photos, receipt, or product information
AI-assisted product identification and understanding
Condition and eligibility assessment
Deterministic underwriting and risk-based pricing
Three personalized coverage options per item
Policy generation and payment / contracting flow
Process
Technical process
- 01URL / Photos / Receipt
- 02Product Understanding
- 03Eligibility & Condition Assessment
- 04Deterministic Underwriting Engine
- 05Risk-Based Pricing
- 063 Personalized Coverage Options
- 07Policy / Payment Flow
Methods
Approach & methods
AI is used to understand messy multimodal product information: identifying the item and supporting condition / eligibility assessment.
Deterministic underwriting logic then calculates eligibility, coverage parameters, deductibles, and pricing from product category, assessed risk, and product characteristics.
The product layer presents three personalized coverage options and continues into policy generation and a payment / contracting flow.
This separation is intentional: the large language model (LLM) does not invent a price; underwriting and business rules remain deterministic.
Contributions
Contributions
Designed an end-to-end applied-AI product flow from item intake to coverage options, policy generation, and payment / contracting.
Separated AI-assisted product understanding from a deterministic underwriting engine so eligibility, deductibles, coverage parameters, and pricing stay rule-based.
Shipped a working public demo and walkthrough of the experience.
Results
Results
Coverage options
3 per item
Personalized combinations of deductible, coverage, and price
Inputs
URL · photos · receipt
Multimodal product information
Underwriting
Deterministic
Rules over generative pricing
Limitations
Limitations
COVER is a product demo, not a licensed insurance carrier or a production insurance company.
The public demo illustrates the workflow; it does not claim live underwriting of real policies at scale.
No dollar prices, conversion rates, or customer counts are reported here because they are not documented as verified metrics.
Lessons
Lessons learned
Multimodal product understanding and underwriting should stay on different sides of the system: one interprets messy inputs, the other calculates structured decisions.
Three clear coverage options are easier to ship and to evaluate than an open-ended generative quote.
Links