Product management · University of Michigan

CookEase

An AI sous-chef that starts with what's already in your fridge

Role: Product & frontend (AI speech frame)
Timeline: EECS 441 · Michigan
Team: 7-person cross-functional team
Product type: Consumer iOS app

What is it

An iOS recipe app that suggests meals from the ingredients you already have, then walks you through cooking out loud so your hands stay free.

Why I built it

We kept running into the same thing while building: most recipe apps assume you have everything on the list. In practice you open the fridge, see what's there, and work backwards. We wanted the app to start where people actually start.

What makes it interesting

The voice assistant does more than read steps aloud. It answers questions mid-cook and suggests substitutions when you're missing something, which is the moment people usually give up and order takeout.

Difficulties & tradeoffs

Usability testing found two problems we hadn't seen: people couldn't find the ingredient list, and almost nobody turned voice command on. We fixed both cheaply. "Done" became "View List," and voice command got a consent prompt instead of a buried toggle. Task success landed at 85–100% after that, and we avoided a redesign we didn't need.

What I built it with

Swift on the client. Django and PostgreSQL on AWS for recipes and pantry state. Computer vision for ingredient recognition, and an LLM layer for rewriting steps and handling substitutions.

SwiftDjangoPostgreSQLComputer VisionLLMAWSFigma

Outcomes

85–100% task success across 8 tested tasks
Two design changes shipped straight out of usability testing

Figma walkthrough & demo

A full walkthrough of the prototype: scanning ingredients, picking a recipe, and cooking with voice guidance.

The process

Interviews first, then competitive analysis, then a value-prop map, then usability testing. Nothing got built unless it traced back to something a person actually told us.

01 — Customer discovery

We interviewed cooks and sorted what they told us into needs, pain points, and wants. Three things came up over and over: people cook from whatever is already in the fridge, vague instructions make them lose confidence halfway through, and nobody trusts their own substitutions.

Affinity map of customer wants and gains

02 — Competitive analysis

We scored five competitors (Whisk, Cookpad, Tasty, Yummly, SideChef) on UI simplicity, voice assistance, ingredient recognition, recipe variety, and how well the front and back end held together. CookEase came out ahead at +12 against +9 for the next best, mostly because everyone else treated voice as an add-on rather than the main way you use the app while cooking.

03 — Value proposition

That gap became the bet. We mapped each pain point to one feature: missing ingredients to scanning and suggestions, unclear instructions to step-by-step voice guidance, and shaky substitutions to real-time help. If a feature didn't map to a pain point, it didn't make the cut.

04 — Usability testing

Six people, eight tasks, clear pass/fail criteria. Two problems showed up: the ingredient list was hard to find and most people missed the voice toggle. Renaming a button and adding a consent prompt fixed both, and task success moved to 85–100%.