Kookie: Personal Cookbook

Kookie is a smart recipe cookbook app with smart grocery integration which allows you to make decision on what recipe to cook on the basis of your preferences and ingredients.

Comment

Onboarding Experience with RevenueCat

https://www.youtube.com/shorts/KGsRif4NpbQ

Inspiration

The inspiration for this app came from my mom, after her passing away what we had was multiple set of diaries and magazines she had noting down the cooking recipes. But I remember one thing vividly that she cooked just 1-2 recipe out of like 200+ recipes she noted. Why was this the case? Because even if she had various recipes noted at a consolidated place, still it was many recipes which slowed her down and she remained stick to staple recipe. That inspired me the idea to make a cookbook app that learns and helps you decide what to cook.

What it does

Kookie is a personalized cookbook app that not only let's you save recipe but also helps you in deciding what to cook by giving daily recommendations learning from what you cooked and saved. It has a Kookie Mode (Decide for me) feature that let's you filter from these recipes you saved based on what you prefer to eat and not only that let's you query the recipe based on what ingredients you have. A smart grocery list which contains the grocery came in from which recipe and also let's you share that list with other people as text.

How we built it

We built it natively for Apple platforms so all platforms from iOS, iPadOS and MacOS are supported. We built it using cursor as the codebuddy and everything is on-device. We have not yet added any AI layer, most of the results are NLP and heuristics based. Core Stack - SwiftUI and SwiftData with RevenueCatSDK for in-app purchase.

Challenges we ran into

The complexity hit me till the last, where I needed to implement an AI layer for completing missing recipe ingredients and instructions. There were multiple approaches to deal with:

  1. Transcribe audio -> processed text -> AI layer -> decodable JSON for display.
  2. Processed text -> on-device LLM/ Cloud LLM -> decodable JSON. as the Cloud LLM would have been very effective in terms of performance and effectiveness. It would have been very expensive. So tried to approach this with on-device LLM wherein tried to first use apple intelligence, but it was not supported in older to iphone 16 devices, so tried to use Ollama to bind a model but this made the app bundle size very big like 800M to 1B parameter models were 400MB and above. So thought of downloading the mode async on first launch and cache it, but the results were not at all promising than expected. There is still AI layer, which will be put, but as were short on deadline, had to go with NLP and heuristics route.

Accomplishments that we're proud of

Made my first app with incredible UI which is appreciated by everyone I demoed my app to. Building this full app end to end in 7 days with a full time job with extra shifts, coded 10hours a day, learnt a lot about paywalls, LLM, on-device LLMs, getting user feedback, learnt about marketing and branding and how important it is.

What we learned

We learned that UX flow and User onboarding is the most important thing for an app. The technical details matter lesser because people want to see thing and they don't care what logic goes in background. Vibe coding is the real thing.

What's next for Kookie: Personal Cookbook

  • I will be implementing the AI layer in a cost effective and performance manner.
  • Implementing more AI functionality to drive engine to give better recommendations based on the time, demographics, occasion.
  • Publishing on app store and marketing.
  • Bringing in community feature to collaborate on app.

Built With

Share this project:

Updates