Chez

• 

  Chez onboarding: see what the app does before you sign up.

• 

  Home screen with your recipes, daily suggestions, and quick actions.

• 

  Import from a link or create from scratch. Paste a TikTok, Instagram, or YouTube URL.

• 

  Your recipe collection with search and filters.

• 

  A recipe imported from TikTok, ready to cook or add to your grocery list.

• 

  Grocery list sorted by aisle. Add ingredients from any recipe with one tap.

• 

  Cook mode: swipe through steps and chat with your AI assistant for substitutions, timing, or technique.

• 

  Ask Chez anything mid-cook. It knows the recipe and your preferences and learnings.

• 

  Weekly cooking challenges with curated recipes from creators.

• 

  A completed challenge recipe with your rating, learnings, and personal version saved.

• 

  Your profile with saved recipes, personal versions, and cooking stats.

• 

  Chef subscription unlocks unlimited imports, 500 messages/day, and recipe memory.

Inspiration

I'm an engineer, but I love cooking every once in a while, and when I do, I like to believe I'm a Michelin-star chef. I research recipes, optimize the flow with the help of AI, execute, and when something works I save it to a personal folder with my notes. It's a great system, but it's duct tape: scattered across tabs, docs, and saved videos that I have to manually stitch together every time.

When I saw Eitan's brief, "From saved recipe to dinner made," I realized most home cooks have the same problem but never get past step one. They save the video and it dies in their bookmarks.

Before Chez: you save a recipe, forget about it, repeat. After Chez: you paste the link, the app walks you through it, and some time later you're plating the dish.

I wanted to build the tool I wish I had, and give creators like Eitan a way to measure real engagement: not just views and saves, but completed meals.

What it does

Someone sees a frozen tomato burrata on TikTok. They paste the link into Chez. The app extracts the full recipe, walks them through it step by step with an AI cooking assistant, and minutes later they're plating the dish. Next time they make it, Chez remembers they preferred balsamic glaze and liked extra basil.

Core flow:

1. Import - Paste a TikTok, Instagram, or YouTube link. Chez extracts ingredients, steps, and timing automatically. Or type a recipe from scratch.
2. Grocery list - One tap adds ingredients sorted by aisle, so you know exactly what to buy.
3. Cook - Swipeable step cards designed for messy hands. Ask the AI anything mid-cook: substitutions, timing, technique.
4. Learn - Chef tier detects when you tweak something ("I added balsamic at the end") and remembers your preferences across every recipe.
5. Complete - Your modifications save as your own version. The original stays untouched.
6. Challenge - Weekly curated recipes from creators push you to try something new.

How I built it

Recipe import fetches video transcripts and runs them through Claude to extract structured data. The cook chat uses RAG: relevant user memories are retrieved via vector similarity and injected into the AI's context, so it actually knows how you cook.

The freemium model gives everyone 20 messages a day and 3 imports a month. Chef unlocks 500 AI messages/day, unlimited imports, and personalized cooking memory. The paywall surfaces naturally after your last free message, not before, so it never blocks you from finishing a cook.

Challenges I ran into

• Reliable extraction from messy transcripts. TikTok transcripts are informal, jump around, and mix instructions with commentary. Getting consistent, structured recipes took heavy prompt iteration plus nonsense detection on both client and server.
• Making memory feel invisible. The learning system had to catch meaningful preferences ("I always add extra garlic") and ignore noise ("I stirred the pot"), without the user ever telling it to save anything.
• Keeping it simple on the surface. The backend handles parsing, AI routing, voice, embeddings, and analytics. The user just sees: paste link, cook, done.

What I learned

• The real value isn't recipe extraction. It's behavior conversion. Getting someone from "I saved this" to "I cooked this" is the product.
• Persistent memory turns a utility into a habit. Users come back because their knowledge lives there.
• Prompt engineering is real product work. The difference between a good and great cooking assistant is largely in prompts and evaluation loops.
• Monetization works best when the paywall shows up at exactly the right moment, not a second before.

What's next

• Hands-free voice controls (wake phrase + next/previous step commands) for messy, real-kitchen moments
• Creator dashboards with conversion rates, completion stats, and top performing recipes
• Cocktails and pastry because the same import, guide, learn loop works for bartenders and bakers
• Pro tools for professional cooks advanced recipe versioning, creative experimentation, and AI-assisted recipe development built on the same engine
• Android release - Built iOS-first for Eitan's TikTok-native audience. The codebase is already cross-platform, Android is next.

Built With

• claude
• deno
• expo.io
• gemini
• openai
• openrouter
• pgvector
• postgresql
• react-native
• react-query
• revenuecat
• supabase
• typescript