OfferScout

OfferScout is your AI shopping savings assistant. Find merchant offers, compare cashback and discounts, check caps, minimum spend and eligibility, and see what you pay and could save before you shop.

Comment

Inspiration

OfferScout started with my own experience shopping at Macy’s. I was considering a purchase but couldn’t easily find an offer in the PayPal app. Only after searching specifically for Macy’s did I discover an 11% cashback offer.

Even then, I wasn’t sure which items qualified or how much cashback I would actually receive. Finding the offer hadn’t answered the questions I needed to make a decision.

That is why I built OfferScout. I wanted shoppers to describe what they’re looking for, discover relevant merchant offers, and understand the conditions and estimated savings in one place.

What it does

OfferScout combines an AI savings assistant with a searchable merchant offer directory. Shoppers can browse offers, filter by footwear category, compare cashback and instant discounts, or ask questions in everyday language.

A shopper can ask “Running shoes under $120” and follow up with “Anything under $80?” OfferScout keeps the relevant search preferences and updates the comparison.

Each sample purchase shows:

  • The amount paid at checkout
  • Any instant discount
  • Estimated cashback
  • Estimated cost after cashback
  • Offer conditions, including caps, minimum spend, expiry and eligibility

A higher cashback percentage does not always mean better value. For example, 15% cashback on a $110 purchase gives only $5 back when capped at $5. A $108 purchase with 10% uncapped cashback gives an estimated $10.80 back, making its estimated cost $97.20. The lower percentage wins in this comparison.

Shoppers can review an offer and approve a PayPal sandbox payment. The receipt separates the completed test payment from simulated estimated cashback.

How I built it

I built OfferScout with TypeScript, React, Vinext, Tailwind CSS and Zod, and deployed it using ChatGPT Sites.

The OpenAI Responses API interprets shopping requests and follow-up questions using structured outputs. Application code then matches offers, checks sample conditions and calculates amounts in integer cents. AI interprets the request, while code handles the money calculations.

PayPal’s sandbox Orders API handles order creation and payment capture. The server checks the order, currency, amount and completed capture status before showing a successful receipt. API credentials stay on the server, and shoppers approve payments themselves.

I used AI development tools to help implement, debug and test the application.

Challenges I ran into

One challenge was making follow-up questions work naturally. Changing a budget should keep the relevant category or merchant, while a new shopping request should start a new search.

Another was keeping the assistant helpful when the catalog could not answer something. It should explain what it cannot verify and offer a useful next step, rather than inventing product details or presenting a sample offer as a confirmed deal.

Keeping cashback separate from an immediate discount was also important. An estimated reward should never make shoppers think the payment itself is lower.

Accomplishments I am proud of

OfferScout brings browsing, conversation, offer comparison and sandbox checkout into one working experience.

The prototype includes 150 fictional footwear merchants and 600 sample offers covering percentage cashback, fixed-dollar cashback, percentage discounts and fixed-dollar discounts.

The current source passes 277 automated tests with mocked provider responses, strict TypeScript checks and a production build. I also tested the live chatbot, completed sandbox purchases and checked the received test payments in the sandbox seller account. The public app and live chatbot were tested in an incognito session.

What I learned

Finding an offer and understanding an offer are two different problems.

A useful recommendation needs to explain the conditions behind the savings. It also needs to distinguish what is known, what is assumed and what the available data cannot establish.

What is next for OfferScout

The next step would be connecting an authorized live offer source, adding richer product information and supporting merchant-specific eligibility.

For this hackathon, OfferScout uses clearly labeled fictional merchants, products and offer terms. Calculations use a fixed October 17, 2026 reference date so the examples remain reproducible. It does not verify current Macy’s offers, other real retailer deals or future Black Friday availability.

All payments use one configured PayPal sandbox seller. No real goods or cashback are issued. Tax is excluded, activation is assumed and customer eligibility is self-declared. Fulfillment and refund handling are outside the prototype’s scope.

OfferScout is an independent hackathon prototype, not an official PayPal product.

Built With

Share this project:

Updates

Submission history