ModoMax: an online store with AI assistants

A full online store I designed and built end to end: catalog, checkout flows, tax documents, a shopping assistant, WhatsApp support and an admin panel with AI tools.

Visit the store →

ModoMax store home page with the headline Protege lo que coleccionas — The deployed store. It runs on test data, so prices, ratings and sales figures are not real.

ModoMax catalog with product cards and filters — Catalog.

ModoMax product page for a wireless charger — Product page.

Decisions worth explaining

Tool calling instead of RAG. The catalog is small and structured, so the assistant searches it through a tool that always returns current data. A vector index would have added moving parts it did not need.
Stale replies are discarded. On WhatsApp, people send several short messages in a row. If a new one arrives while the model is working, the old reply is dropped rather than answering a question that has already changed.
The public endpoint is limited. The assistant needs no login, so it is limited to 10 requests per minute per IP address.
AI output is validated. Business insights in the admin panel must pass Zod schemas before they are shown.

Stack and tests

TypeScript, Next.js, React, Prisma, PostgreSQL, Vercel AI SDK and the Anthropic API, with 48 test files across Vitest and Playwright.

The source is private. Happy to walk through it in an interview.