What I Made

The product question

A meal can leave a photograph behind while losing the details that would make it worth revisiting: what went into it, how it tasted, and what to change next time. Recording everything in a form adds effort at the moment someone wants to eat.

What I Made explores a lighter approach to keeping a cooking history. My role is product design and hands-on prototyping. It is built for home cooks who want to remember individual attempts and discover patterns across the dishes they make.

The experience

Capture starts with a photograph and dish name. A cook can add a natural voice note, review the suggested fields, and save without accepting every suggestion. One occasion can contain several dishes and photographs.

The same archive supports three perspectives: a chronological Journal, a photographic Year view, and a culinary Map. Dish histories connect repeat attempts; filters help rediscover a meal. Recipe Ideas can be imported or researched when requested, then saved as editable snapshots. Backup and restore give the owner a way to retain the archive outside the browser.

How it works

1. 01
   Capture on the device

   The PWA processes photographs locally and accepts a dish name, typed notes, or speech.

2. 02
   Request help when useful

   Authenticated cloud services transcribe speech or propose structured details from submitted text.

3. 03
   Review and save locally

   The cook confirms the entry; IndexedDB stores the occasion, dishes, attempts, and photographs.

4. 04
   Rediscover the history

   Journal, Year, and Map derive their views from the same account-scoped archive.

Cloud assistance is optional. The public sample reads a separate fictional dataset and does not open a private archive.

Cognito establishes who may enter the private app; it does not store the cooking archive. Network services receive the bounded inputs needed for a requested action. Personal photographs stay on the device. A separate metadata analytics path can record allowlisted activity without receiving meal records or images.

Stack and choices

App and archiveJavaScript PWA · IndexedDB

Keep capture, photo processing, and the source-of-truth archive on the device.

Access and servicesCognito · API Gateway · Lambda

Separate invitation-only membership from bounded cloud assistance.

AssistanceAmazon Transcribe · Bedrock

Turn speech into text and propose editable capture or recipe details.

DeliveryAWS Amplify · Service worker

Deliver the installable app and preserve useful local access when connectivity drops.

Decisions and design lessons

Capture effort is a product constraint. Photo and name form a minimal path. More capable assistance is useful only if it reduces effort without making the cook correct a long, unexpected form.

Suggestions should remain suggestions. The person can inspect and edit AI-proposed details. Deterministic local matching and country resolution handle tasks that do not need a model’s judgment.

Local ownership comes with responsibilities. Device-local storage protects the archive from routine cloud upload, but sign-in alone does not provide cross-device backup. Export and validated restore are part of the product, not an afterthought.

Current state and next step

The invitation-only PWA has a publicly accessible fictional sample. Repository records describe owner iPhone validation and client/service checks, with remaining device checks documented separately. A public sample demonstrates the experience without exposing a private cooking history.

A useful next question is whether repeat cooks can recover a specific past attempt quickly, especially as dish names and occasions accumulate. The public repository includes the product requirements, architecture, and media attribution.