Milestone 0
Project shell
React, TypeScript, Vite, and Tailwind setup with a bank-insurance layout, sidebar navigation, header, and placeholder pages.
AI-assisted claims review
CardCover
A staged prototype for reviewing card-linked travel insurance claims. The standalone app starts as an operational shell, then grows into a human-in-the-loop review tool for baggage, flight disruption, and personal item claims.
Open the prototype to explore the queue, selected-claim workbench, AI recommendation evidence, challenge panel, stress tests, decision form, and audit log.
Open standalone app →Why CardCover exists
CardCover is an AI-assisted travel insurance claims review tool for bank credit card holders. It is designed for insurance reviewers who need to assess whether a card-linked travel claim is eligible, complete, policy-aligned, and supported by reliable evidence before a human decision is made.
The product focuses on common travel insurance claims connected to premium credit card benefits, where reviewers must balance customer experience, policy limits, fraud risk, and auditability.
Supported claim scenarios
- Lost baggage
- Delayed baggage
- Flight cancellation
- Flight delay or missed connection
- Stolen or damaged travel items
Reviewer questions
- Is the cardholder eligible?
- Was the trip paid with the right credit card?
- Is the claim type covered?
- Are the required documents complete?
- Does the timeline make sense?
- Is the amount within the policy limit?
- What evidence supports the AI recommendation?
- What does the human reviewer still need to verify?
Prototype focus
CardCover explores how claims teams could combine structured policy checks, AI recommendations, evidence gaps, risk flags, and human oversight in one review environment. The current build establishes the React application shell, synthetic data model, claims queue dashboard, a three-column claim review workbench, deterministic mock AI review explanations, a Challenge AI review assistant panel, claim stress-test checks, a human reviewer decision form, an in-session audit log, and a portfolio-ready accessibility polish pass.
Milestone 1
Claim data model
Typed synthetic data for five supported travel claim types, including documents, risk, recommendations, policy matches, and oversight tasks.
Milestone 2
Claims queue
A dashboard table with calculated queue metrics, readable risk/status badges, evidence status, and row selection into the Review Workbench.
Milestone 3
Claim workbench
A three-column selected-claim review layout with summary, evidence, missing evidence, policy matches, timeline, risk flags, and oversight checklist.
Milestone 4
Mock AI review logic
Deterministic, API-free recommendation explanations framed as decision support for human reviewers.
Milestone 5
Challenge AI panel
A review assistant panel with deterministic claim-specific responses for evidence, policy, escalation, and customer explanation checks.
Milestone 6
Claim stress test
Deterministic pass, needs-review, and fail checks for eligibility, coverage, evidence, timeline, amount, missing data, consistency, and oversight.
Milestone 7
Reviewer decision form
A required-note human decision workflow that updates claim status while keeping the AI recommendation visually separate.
Milestone 8
Audit log
An in-session log that records claim opens, AI review events, challenge questions, customer explanations, escalations, and reviewer decisions.
Milestone 9
Polish and accessibility
A calmer B2B interface pass with stronger focus states, clearer labels, responsive review signals, improved empty states, and better table semantics.
Current scope
The prototype uses synthetic data only. It has no authentication, no external APIs, and no live claim decisions. Detailed document review, reviewer notes, policy reasoning, and final decision actions are reserved for later milestones.