Almond v1

Almond was built to save a recipe from wherever it was found, make it readable, then keep it useful while cooking.

I designed and built the native Swift app and the backend AI creation pipeline that runs on Cloudflare Workers.

Recipes can be created in serveral ways. A user could paste a website link, upload an image(s) from the photo library, capture a photo of a recipe using the camera, or share from another app.

Each path resolves into the same clean recipe structure. The pipeline uses an AI agent that looks for metadata and selects the best path forward to create the recipe.

The recipe library is the main home of the app. It used image-led recipe cards so saved recipes could be recognized quickly, with enough metadata to separate similar items without turning the view into a database.

Recipe search is a core part of the app, powered by hybrid vector search. This allows recipes to be found using ingredients, instructions, source text, or even partial names—leveraging both semantic understanding and traditional text matching.

Collections allow users a way to group recipes by meal, occasion, person, or project without turning organization into a filing task.

The recipe view presents a minimal, easy-to-scan layout perfect for keeping open while you cook. Key details like ingredients and numbered steps are clearly prioritised.

The share extension made Almond available from the places people already found recipes. A user could share a URL or image into Almond, close the sheet, and let the main app create the recipe.

Built with a custom Metal shader to show some personal touches of Almond absorbing the recipe.