Inspiration
Millions of home sellers, like bakers, tailors and craft makers, run their business on WhatsApp and Instagram. They type every product detail by hand, answer the same questions again and again ("Is it eggless?", "What size?"), and chase payments. Group gifts are even harder: one friend pays for everything, then chases the others for their share.
What it does
Kadai is a shop in a chat.
- Seller: uploads one product photo and an optional note. Google Gemini looks at the photo and writes the name, description, a suggested price and the details. The seller reviews, edits and publishes.
- Buyer: chats with an AI shop assistant that answers only from the product details. If it doesn't know, it says the seller will confirm. It refuses discounts and keeps the price fixed. When the buyer says "I'll take it", a Confirm & pay button appears and they pay with PayPal.
- Split with friends: the buyer adds friends' names and Kadai splits the total exactly, to the cent. Everyone gets a shared tracker, each friend pays only their own share through PayPal, and AI writes a friendly reminder naming only the people who haven't paid yet.
- Seller dashboard: every order shows as Pending, Paid, Partially paid or Fully paid.
Try it:
- Shop: https://kadai-copy.onrender.com/shop
- Seller dashboard: https://kadai-copy.onrender.com/seller
- Add a product with AI: https://kadai-copy.onrender.com/seller/new
Payments run in PayPal sandbox (test mode). Free hosting, so the first visit can take up to a minute to wake up.
How we built it
- Next.js + TypeScript, Tailwind CSS and shadcn/ui
- Vercel AI SDK with Google Gemini Flash (reads photos, returns structured JSON) and Groq (Llama 3.3) as a text fallback
- PayPal Orders API v2 (sandbox): create order → buyer approves → server captures. For splits, a share's PayPal order is created only when that friend taps "Pay my share", so links never expire while friends take their time.
- Supabase for the database and photo storage, with row-level security on
- Render free hosting, public GitHub repo under the MIT license
Safety rules: the server always reads the price from the database (never the browser), all money maths is done in cents, and no secrets are in browser code.
Challenges we ran into
- Splitting money exactly: $22 between 3 people is $7.34 + $7.33 + $7.33, never $7.333.
- Making payment pages safe to refresh without charging twice.
- Keeping the AI honest: no invented ingredients or allergy facts, no discounts, and ignoring prompts that try to change the price.
- Sandbox setup: USD payments need a US sandbox merchant account.
Accomplishments that we're proud of
A complete shop-to-payment flow, with AI listing, an AI shop assistant, PayPal checkout and group split payments, working live, built using only free tools.
What we learned
How the PayPal Orders API create–approve–capture flow works end to end, how to get reliable structured output from a vision model, and how much safety work an AI that talks about prices and payments needs.
What's next for Kadai – Shop in a Chat
Seller login and multiple shops, sharing products straight to WhatsApp, delivery tracking, and more currencies.
Built With
- google-gemini
- groq
- nextjs
- paypal
- render
- shadcn-ui
- supabase
- tailwindcss
- typescript
- vercel-ai-sdk