Inspiration
Ever had a little extra money and wondered what to spend it on? Picking takes time, and it's easy to overspend. We wanted an AI that does the shopping for you without ever going past a budget you set.
What it does
PuffrPay is an AI shopping agent that spends your fun money inside rules you set: your fun money, on a leash.
- Just ask: tell Puffr something like "a birthday gift for my sister who climbs, under $50" and it finds real products, then shows you up to 5 picks.
- Rules decide every purchase: each item is checked against your monthly cap, per-purchase max, auto-approve limit and blocked categories:
- ✅ Allow: small purchases go straight through
- ✋ Ask: bigger ones wait in your approval queue
- ⛔ Block: anything breaking a rule is stopped before you can pick it
- Pays with PayPal: approved purchases are paid for with PayPal (sandbox).
- Activity dashboard: an AG Studio report with spending vs. your cap, spend by day, a full purchase log in AG Grid, and the Leash report, which shows everything the rules stopped.
- Ask about your spending: questions like "what got blocked this week?" filter the dashboard for you.
- Puffr the pufferfish: our mascot puffs up as you get closer to your monthly cap, so you can see your budget at a glance.
How we built it
- Next.js + React + TypeScript for the app, with PostgreSQL storing purchases, rules and chat history
- Google Gemini plans the search and chooses picks, with an offline mock mode for testing
- Channel3 for live product search
- PayPal Orders API (sandbox) to create, approve and complete each payment
- AG Studio + AG Grid for the Activity dashboard, custom chart widgets and the AI assistant panel
- A rules engine that checks every purchase in a fixed order: blocked category → per-purchase max → monthly cap → auto-approve, otherwise ask. It's covered by unit tests.
Challenges we ran into
- UI/UX: making a money app feel fun and trustworthy, with a dark/light glass design that still feels fast
- AG Grid & AG Studio: fitting a full interactive dashboard (custom widgets, cross-filtering, saved layouts) into our design without breaking how it works
- Channel3: turning messy real product data into clean picks the rules could check reliably
- Keeping the AI honest: making sure the AI can only suggest products from the search results, and that the rules always have the final say
Accomplishments that we're proud of
- The UI/UX, especially the Activity dashboard
- A full PayPal payment flow, from approval to completed purchase
- An AI that is genuinely helpful but can never spend past your rules
What we learned
- A lot about working with APIs (PayPal, Channel3, Gemini) and handling their failures gracefully
- How to design a website around what users actually need: clear choices, clear limits, no surprises
- That performance is part of UX: smoother animations and lighter visual effects made the dashboard feel much better
What's next for PuffrPay
- 📱 A mobile app with a home-screen widget: tap it and ask Puffr for gift ideas or "what should I buy with my $20 left?"
- 🔔 Notifications for purchases waiting on your approval
- 👥 Shared budgets, like a parent setting the rules for a teen's allowance
Built With
- ag-grid
- ag-studio
- channel3
- claude
- cognition
- devin
- gemini
- next.js
- node.js
- paypal
- postgresql
- react
- typescript
- vitest