Paywise

Inspiration

Your laptop stops working 14 months after you bought it, two months after the manufacturer’s warranty expired. But the credit card you used may have extended that warranty by another year.

Benefits like extended warranties and theft protection are easy to miss when they’re buried in a long PDF. Discover, an American financial institution, even discontinued its extended warranty and purchase protection benefits in 2018, citing “prolonged low usage.”

We built Paywise to show you which benefits apply before you pay, then help you use them if something goes wrong.

What it does

Paywise on an iPhone: double-tap at checkout, the screenshot is redacted on the phone, the answer appears over the checkout, the purchase is saved, and months later "Can I claim?" says may be eligible.

Tell Paywise what matters most when choosing a card: money, service, or something else you describe yourself.

At checkout, double-tap the back of your iPhone. Paywise reads what you’re buying and compares your cards’ cash back and purchase protection, taking your preference into account. One tap saves the purchase and its coverage dates.

If the item later breaks or gets stolen, ask, “Can I claim?” Paywise checks the terms, explains whether you may be eligible, and lists the documents and deadlines. AI helps you draft the claim for you to review and submit.

Your screenshot stays on your phone. Paywise extracts the text and removes card numbers, names, and addresses before sending it for analysis.

How we built it

We prioritized simplicity, scalability, and generalizability in our system architecture. We developed a native iOS app that captures checkout screenshots, processes them through our backend, and delivers personalized credit card recommendations directly through Apple's Dynamic Island in 0.3s.

Our backend maintains a comprehensive credit card rewards database, transforms unstructured screenshots into structured JSON, and leverages Jev for intelligent, policy-driven reward optimization. We automatically refresh reward policies every 48 hours while adopting a privacy-first architecture that keeps sensitive financial data off our servers.

Challenges we ran into

We primarily used Jev from Typesafe alongside a VLM, improving accuracy by 5% and reducing runtime by 95% compared with VLM-only approach.
Making the back tap setup smooth and ensuring the back-tap shortcut worked reliably without opening our app.
Recommending the best card for each purchase without overlooking a user's existing card benefits for marginal cashback gains.
Grounding every answer in the corresponding card's official benefits guide to ensure accuracy and traceability.

What we learned

Cash back is easy to compare. Purchase protection takes more work: it depends on the item, the card’s terms, and when something goes wrong.

We also learned to be precise about uncertainty. “May be eligible,” backed by a specific clause and clear next steps, is more useful than a confident answer we can’t support.

What's next for Paywise

We’ll support more credit cards and gather user feedback to improve the app’s navigation and screen design. As Paywise grows, we’re considering a seven-day free trial followed by a paid subscription, giving users time to try the card recommendations and AI claim assistant before subscribing.