Cook Master

• 

  Icon

Inspiration

When I ran into Eitan's brief something sparked enough joy for me to take a look at the project from a new angle. With the inflection point that Whisper models and AI in general brought, I thought we could craft something that works with you at all times, like a command center for your kitchen. Bringing everything that makes a recipe/cooking app nice, and just cranking the knob a little bit more, so even before you get to the kitchen you hit a flow state, the week has been planned, the shopping list is ready and by the time you're home, your headspace is free.

What It Does

Cook Master becomes an all-in-one command center for the kitchen and the family.

It allows weekly meal planning based on calories, dietary preferences, and what is already sitting in the pantry, then seamlessly generates shopping lists for only what is missing. It surfaces recipes in the "Cook With What You Have" feature to reduce food waste and unnecessary trips to the store.

The Quick Cook feature lets users jump straight into cooking by selecting from their pantry, shopping list, or personal preferences, or by simply snapping a picture of whatever ingredients are on the counter.

Social media imports are where Cook Master truly shines. The app has a robust pipeline for importing TikTok and Instagram videos. Users can import without leaving the social media app via the Share Extension. No written description on the video? No problem. We analyze the video, extract step by step instructions and ingredients, and deliver a clean, structured recipe card. All of this happens in a cost effective way, making the app not just a showcase but genuinely production ready.

Hands free cooking mode with voice commands means users can say "next step" or "repeat that" without ever touching their phone with messy hands.

And the cookbook scanner lets users digitize their grandmother's recipe book, page by page, using on device OCR and AI powered extraction.

How We Built It

The app is built natively in Swift using a SwiftUI and UIKit hybrid architecture with the MVVM pattern and a dedicated service layer. The backend runs on Parse Platform, giving us a flexible NoSQL database that scales without ops overhead. Subscriptions are managed through RevenueCat, and AI features are powered by the OpenAI.

For the social media import pipeline, we built a custom URL resolver using WebKit that follows TikTok and Instagram redirects, then passes the resolved content to our SvelteKit backend hosted on Vercel. The transcription and extraction pipeline was designed from the ground up to minimize API calls while maximizing accuracy and cost.

On the device side, we leaned heavily into native Apple frameworks: Vision for OCR, Speech for voice commands, AVFoundation for camera based scanning, and PhotosUI for media selection. The goal was to keep the dependency footprint small and the experience fast.

Challenges We Ran Into

The social media import pipeline was by far the hardest piece. TikTok URLs go through multiple redirects, and the structure of the final page changes frequently. Building a resolver that works reliably across short links, regional variants, and private videos required creative problem solving and a lot of edge case handling. Making it also as cost effective as possible, fast and reliable and ensuring it will never be blocked or bloat the users device was prioritized. Everything is cloud based, ensuring security and easy rate limiting.

Another major challenge was state synchronization between the shopping list, pantry, and meal planner. When a user checks off an item on the shopping list, it should optionally move to the pantry. When a meal plan generates a shopping list, it should not duplicate items already in the pantry. Getting these interactions right without race conditions required careful use of Combine publishers and actor based concurrency with Swift's async/await model.

Finally, keeping the AI features cost effective at scale was a constant design constraint. We had to architect the system so that it could serve real users at real volumes without the bill becoming unsustainable and ensuring the app has great margins.

Accomplishments That We Are Proud Of

Cost effectiveness is the accomplishment we keep coming back to. It is built to run in production and work with the monetization plan.

The advanced caching system for social media imports is something we are especially proud of. Cooking videos on TikTok/IG regularly garner millions of views. We built a system that only needs to process a video once. If another user requests the same video, they get the recipe instantly at zero additional cost to us. At scale, this turns what could be an expensive feature into an incredibly efficient one.

Everything was built with sustainability and scalability in mind, it can support large traffic spikes. The app can now realistically handle millions of users even as it stands.

What We Learned

I would say mainly, re-learned that the hardest part of an AI powered app is not the AI. It is everything around it: the state management, the caching strategy, the edge cases in third party platform integrations, the UX that makes complex features feel simple and cost effective. The AI is the engine, but the chassis matters just as much.

Again always re-learning to resist over engineering. Building with a robust architecture in mind, stripping things back, using native frameworks wherever possible, keeping the dependency list short, and letting the architecture serve the user instead of the other way around. Everything is built with modularity and millions of potential users in mind.

What Is Next for Cook Master

Personalized recipe surfacing based on user behavior and preferences. All the elements are already in place, so we can just implement it.

The next step is building a taste profile that acts like a fingerprint. The more a user interacts with the app, the better the recommendations become. Think of it as a recommendation engine that understands not just "you like Italian food" but "you prefer 30 minute weeknight meals with minimal cleanup that lean toward Mediterranean flavors." This can be used to surface meals during lunch break when at work, build a shopping list that the user can approve fast.

Also plan to expand the social layer, letting families and roommates share pantries, coordinate shopping lists, and plan meals together in real time. Maybe even community surfacing with "what's hot" around so users can cook with local ingredients and recipes.

Cook Master started as a frustration with lost screenshots and forgotten grocery lists and my personal dietary restrictions while being Celiac. It is becoming the kitchen companion I always wished existed.

An advanced affiliate system has been planned which will help with distribution greatly. Our import system is robust now and can be expanded to import from any source easily. YouTube and website imports are also right around the corner, same strategy as the TikTok and Instagram built in efficiently, without ai web-search which would hurt margins.

Siri Intents & Widgets so the app works seamlessly across the OS.

Built With

• applespeech
• applevision
• avfoundation
• parse
• revenuecat
• sveltekit
• swift
• swiftui
• vercel