← Carlos Cardona

SUPERMENU. Smart system for restaurants.

Dec 2022 - Dec 2024 · supermenu.app · Founder & CEO

---

SuperMenu is a smart system for restaurants: using AI to cut the busywork and lift profitability. Two products. An interactive menu where every dish has a photo, ingredients and filters, with a personalized suggestion engine so each guest finds what fits them. And Marketing +AI (coming): private feedback, online reputation and loyalty that turn first-timers into regulars.

A diner opens El Rincón de Vespok's menu, picks a dietary preference, flags allergies from the 14-allergen grid, and lands on a menu built for them: dishes by category, real photos, prices, search, filters. The restaurant's brand is everywhere; ours is nowhere. Two Madrid locations ran on it.

INSIGHT

Eating out is personal. Everyone arrives with preferences, a budget, sometimes an allergy, and they want to see what they're about to order instead of decoding a dated text menu. All of that belongs to the dish, not the restaurant. Google Maps finds you a place; nothing helped you find the plate that actually fits you.

WHAT I BUILT

  • Interactive photo menu
  • Budget, allergen and diet filters
  • Diner preference onboarding
  • Personalized dish suggestion engine
  • Multi-location restaurant back office
  • Live menu design customization
  • Visits and reviews analytics

[Images / demo clip go here]

HOW IT'S BUILT

  • Diner PWA and admin panel, both Vue 3
  • Node and Express API, MongoDB, Firebase auth
  • Multi-tenant: each restaurant on its own white-label domain
  • Recommender: a Python service, collaborative filtering, rate-five cold-start, distance penalty
  • In-house dish search on MongoDB, after dropping a paid search vendor
  • Media pipeline: upload, resize, serve from S3
  • Anonymous diner sessions, so guests scan and browse with no signup

DECISIONS

Modeled the dish, not the restaurant, as the core entity. A restaurant has no price; a dish does, so a guest can say "I have 20 euros" and hide everything above it. A dish has ingredients, so you can filter by allergy and diet. A dish has a photo, so people see what actually arrives. That one modeling choice unlocked the whole product. Its cost: every dish must be photographed and structured, an enormous data burden that later became the reason restaurants never fully adopted it.

Made it fully white-label, because restaurants asked. An earlier version imposed our look, and owners pushed back: they wanted their own colors, typography and brand. So we rebuilt the menu as white-label, each restaurant on its own domain with SuperMenu invisible, and added the onboarding and splash screen they specifically requested. The ones who tried it liked it, because it felt like theirs, and it was the part they actually used. Cost: we built no brand or network effect of our own.

Pivoted marketplace to SaaS by reasoning, not by asking. A two-sided marketplace needs demand and supply at once, and we lacked the resources, so we turned what we'd built into a tool for individual restaurants. We made that call in a room, not with customers. The original problem was never that restaurants needed a better menu. We invented a problem to fit the technology we had.

Built the recommender as a separate, data-hungry service. Collaborative filtering needs volume, and it was a sound design for a city marketplace with 1,000+ dishes and many ratings. The SaaS pivot moved it into single-restaurant scope, which starved it: one restaurant has neither the dishes nor the ratings to train or run it. The pivot didn't just change go-to-market, it invalidated the core technical asset, and it never launched.

OUTCOMES

  • A fully built product: diner app, admin, and a real recommender
  • About 10 restaurants onboarded, none paying or actively using it

LEARNINGS

  • Watch what customers do, not what they say. Ours called it wonderful and never acted on it. "They like it" is not adoption. We talked to them plenty; the miss was trusting words over behavior. Build for what people will actually act on, not what they praise.
  • Choose your battles. Selling to consumers, or to small restaurants that behave like them, is brutal, and a market everyone is already fighting over is a bad default unless you have a real edge.
  • Go-to-market can't be done part-time. A hard market needs full-time selling, and we didn't put full-time hours behind it.

STATUS

Finished. We stopped work at the end of 2024 to focus on another project. The product is still online but frozen: not maintained, not sold, and with no plans to restart.