Miso

Your all-in-one cooking companion, save recipes from anywhere, track your pantry, plan meals, and cook step-by-step, all powered by AI.

Comment

Inspiration

We love cooking, but the experience around it is broken. Recipes are scattered across YouTube, TikTok, Instagram, and random blogs. Grocery lists live in notes apps. Meal planning happens in our heads. And half the time we open the fridge and still don't know what to make.

We wanted one app that ties the entire cooking journey together, from discovering a recipe to putting food on the table. That's Miso.

What it does

Miso is a mobile cooking companion that brings your entire food life into one place:

  • Save recipes from anywhere - paste a link from YouTube, TikTok, Instagram, or any recipe URL and Miso extracts the full recipe automatically using AI.
  • Smart pantry - scan receipts, photograph your fridge, or snap product barcodes. Miso tracks what you have at home.
  • Meal planning - drag recipes from your collections into a weekly calendar. Plan meals in seconds.
  • Auto shopping list - when you save a recipe and you're missing ingredients, Miso adds them to your shopping list. Check items off and they move to your pantry.
  • Interactive cooking mode - step-by-step guided cooking with timers, serving adjustments, and progress tracking.
  • AI agent - a conversational kitchen co-pilot that can import recipes, scan receipts, generate new dishes from your pantry, and plan your week.
  • World map - every recipe can be pinned to a location. Explore dishes from around the globe.
  • Social feed - discover trending video recipes, follow creators, share your own dishes, and build a cooking community.
  • Achievements & streaks - gamified cooking progress that keeps you motivated.
  • Miso Plus (RevenueCat) - premium subscription unlocking the AI agent, unlimited collections, and expanded recipe saves.

How we built it

Mobile app: React Native with Expo, using Nativewind (TailwindCSS) for styling. Navigation with Expo Router. State management with Zustand and React Query.

Backend: Hono.js running on Vercel serverless functions. PostgreSQL on Supabase with Prisma ORM. pgvector for recipe similarity search and AI-powered recommendations.

AI pipeline: OpenAI GPT-5-mini via Mastra AI agent framework. Tools for recipe extraction, receipt scanning, meal planning, and pantry management. BrightData for scraping recipe data from social platforms (YouTube, TikTok, Instagram).

Media: Cloudinary for image and video storage. Pexels for stock video content.

Monetisation: RevenueCat handles subscription management for Miso Plus. Free tier gives core functionality; Plus unlocks AI features, unlimited collections, and priority support.

Infrastructure: Vercel for deployment, Supabase for database and auth, GitHub for version control.

Challenges we faced

  • Recipe extraction from social media - TikTok, Instagram, and YouTube all have different content formats. Getting reliable recipe data from video captions, transcripts, and page metadata required building platform-specific scrapers with BrightData and falling back to AI extraction when structured data wasn't available.

  • Pantry intelligence - connecting the dots between what's in your pantry, what a recipe needs, and what goes on your shopping list sounds simple but required careful fuzzy matching logic to handle ingredient name variations (e.g., "chicken thighs" vs "boneless chicken").

  • Real-time cooking mode - building a step-by-step cooking experience that feels responsive and doesn't lose your place required thoughtful state management and progress persistence.

  • RevenueCat integration - implementing a clean subscription flow that gates AI features while keeping the free experience valuable took several iterations to get right.

  • Keeping it simple - the biggest challenge was scope. Miso does a lot, and keeping the UX clean and not overwhelming was a constant design conversation.

What we learned

  • AI is incredibly powerful for recipe extraction, but it needs guardrails. Structured prompts and validation layers were essential to get consistent results.
  • Fuzzy matching for food items is a harder problem than it looks — plurals, abbreviations, regional names, and unit conversions all need handling.
  • RevenueCat made subscription management dramatically simpler than building it from scratch. The webhook integration for server-side entitlement checks was clean and reliable.
  • Building for mobile-first forced us to think about every interaction in terms of speed and thumb-reachability.

What's next for Miso

  • Smart suggestions - proactive recipe recommendations based on what's expiring in your pantry.
  • Collaborative collections - share collections with family or housemates and plan meals together.
  • Nutrition tracking - detailed macro and micronutrient breakdowns per recipe and meal plan.
  • Offline mode - full cooking mode access without internet.
  • Voice control - hands-free cooking with voice commands for timers and step navigation.
  • Bug Fixes - Fix all bugs related to the app currently from the AI to the basic functionality

Built With

Share this project:

Updates