Personal Project

RecipeAssist — save any
recipe, cook from
anywhere.

An AI-powered recipe book I designed and built for how families actually collect recipes. Paste a link, drop a YouTube video, snap a photo, or just describe a dish — an AI pipeline turns it into a clean, structured recipe, translates it if needed, suggests ingredient substitutions, and files it in a shared library you can search and cook from. Live at recipeassist.app.

Visit recipeassist.app
Role
Designer · Engineer · PM
Timeline
2025 – present
Platform
Web — recipeassist.app
Scope
AI recipe manager
RecipeAssist browse view — click to play the demo video
The Project

A recipe book that does the tedious part for you.

Every family's recipes live scattered across screenshots, bookmarked blogs, YouTube videos, and texts from relatives — and the good ones get lost. RecipeAssist is my answer: a single, shared, AI-powered cookbook. You add a recipe however you found it — a URL, a YouTube link, a photo of a cookbook page, or a few words — and it comes back clean, structured, and searchable, with the original source kept intact.

I designed and built the whole product solo — the interface, the multi-agent AI pipeline, the data model, and the infrastructure — using AI-assisted vibecoding to take it from a PRD to a real app running in production for my family.

How it works

Add it any way you have it. Get back something clean.

Adding a recipe is a conversation. You hand RecipeAssist whatever you've got, and a chain of specialized AI agents figures out what you meant, extracts the recipe, and files it — no forms, no copy-pasting ingredients line by line.

The Add Recipe assistant — paste a URL, a YouTube link, upload a photo, or describe a recipe in your own words
01 · Capture
Add it any way
Paste a recipe URL, drop a YouTube link, upload a photo of a page or screenshot, or just describe a dish in your own words. Four inputs, one box.
02 · Structure
AI extracts the recipe
The pipeline scrapes the source, reads the photo, or pulls the video transcript, then extracts clean ingredients, steps, and tags — keeping a link back to the original.
03 · Translate & adapt
Any language, any pantry
Found the perfect dish on a foreign-language site or video? RecipeAssist translates it into a clean, consistent recipe — and suggests ingredient substitutions when you're missing something or want to adapt it.
04 · Find
Search & browse
A card view of the whole collection with semantic search and filters by cuisine and main ingredient — ask for "something with chicken and rice" and get it.
05 · Share
A shared family book
Add friends and family so the cookbook is collaborative — everyone contributes, every recipe is tagged by who added it, and nobody loses the good ones.
Tech Stack

A multi-agent AI system on a modern serverless stack.

The core is a chain of specialized agents rather than one big prompt: a classifier routes each message, then dedicated agents store, search, generate, or translate. It runs on a serverless Cloudflare backend so a single person can operate it at real scale.

Multi-agent AI pipelineAn intent classifier routes each request to purpose-built agents — store, search, generate, translate, and chat — powered by OpenAI models.
Next.js 14 + React (MUI)App Router front end with a Material UI design system, React Query for data, and Framer Motion for the interactions.
Cloudflare (D1 · R2 · Vectorize)Deployed on Cloudflare via OpenNext — D1 for the database, R2 for recipe photos, and Vectorize for embeddings.
Semantic searchRecipes are embedded into a vector index so search understands meaning — matching by ingredient, cuisine, and intent, not just keywords.
Ingest & visionCheerio scrapes recipe sites, a YouTube transcript pipeline reads videos, and Sharp / HEIC conversion prep photos for OCR extraction.
Auth, limits & opsSupabase auth, Upstash Redis rate limiting, Resend for transactional email, and PostHog for product analytics.