Project 04 — 2024 · Unshipped experiment

MealTrip

Meal planning for every scenario at once. It got out of hand.

The idea started off reasonably. I wanted to make a group food planner that was more fun than a spreadsheet. I had just downloaded Claude and started jamming; one poorly defined prompt after another. It was so fun. Add a new feature. Animate some things that nobody expects to be animated. Pack the delightful interface elements in everywhere. It became one planner that could handle any eating situation — weeknights, dinner parties, road trips, backpacking, restrictions, budgets. Confetti. Animal sounds. Gamification for no good reason.Every scenario I added made the data model more general and the interface less usable, until the thing could describe any meal plan and help with none of them. I stopped before launch. Vittles is what that lesson turned into: a few scenarios taken seriously, and more niche. This project taught me an important lesson about scope, and that planning is more important than ever when you are using a generative model to build a product.

The problem

I got caught up in the newbie phase of ai magic. I chased novelty instead of product definition. I was so amazed that I could build things that would have previously taken me a week in a hour or two.

Product decisions

The useful decision was the one to stop. This project was part of the journey, a learning experience, not a product.

Tech

React, Tailwind, NextJS - deployed on Vercel

React · TypeScript · Node

Landing — one planner, pitched at every eating situation at once
Events — trips and parties share one model, which is where the trouble started
My Kitchen — a recipe library that exists independently of any trip
Dashboard — dietary compatibility across the library, reported before anything was planned