CartMind

Inspiration Online shopping means endless scrolling and filtering. Shoppers already know what they need ("a gift for a coffee-loving friend under $60") — they just can't say it to a store. We wanted a store you can talk to, with a checkout as quick as the question.

What it does You type a need and an optional budget. AI reads the whole catalog and returns 1–4 picks, each with a personal reason, plus a summary. The server validates every pick and re-checks the budget, so the AI can't invent products or overspend. Add picks to the cart in one click and pay with PayPal Smart Buttons. After approval the payment is captured and a receipt shows the PayPal order ID, capture ID and status. How we built it Frontend: React 19 + TanStack Start + Tailwind CSS v4. AI: OpenAI gpt-6-astra through the Lovable AI Gateway, called with the Vercel AI SDK. The model gets the catalog + request and returns JSON picks. Payments: PayPal JavaScript SDK renders Smart Buttons. Our server gets an OAuth token, creates the order with the Orders v2 API (prices recomputed from the catalog, never trusted from the browser), and captures it after approval. Hosting: server functions on Cloudflare Workers; built with Lovable. Challenges Keeping AI output trustworthy: we validate IDs and enforce the budget server-side. Making sure checkout amounts can't be tampered with from the browser. Wiring PayPal's client-side approval to server-side create/capture cleanly. What we learned How to combine an AI "decision" step with a secure payment step, and how PayPal's Orders v2 create → approve → capture flow fits a modern server-function app.

What's next Real merchant catalogs, PayPal Pay Later messaging, saved preferences, and multi-turn chat refinement ("cheaper", "more eco-friendly").

Built With

• cloudflare-workers
• lovable
• lovable-ai
• openai
• paypal
• paypal-javascript-sdk
• paypal-orders-api
• paypal-sandbox
• react
• tailwindcss
• tanstack-start
• typescript
• vercel-ai-sdk
• vite