Umami

Inspiration

As a university student, we hit a breaking point when meal planning felt more exhausting than the cooking itself. we'd skip meals, eat irregularly, and watch our classmates do the same - one friend even developed a stomach ulcer from neglecting regular meals. That's when it clicked: we were sacrificing our health for our GPAs, which is completely backward. Our bodies need consistent fuel to function, both physically and mentally. We wanted to create something that felt less like a chore and more like a comforting companion. That's where Studio Ghibli came in - those films make even mundane tasks feel warm and magical. And Umami? It's that deep, satisfying flavor that makes food worth savoring - a fifth sense of sorts. The name represents what we wanted with this app to be: your personal kitchen assistant that brings back the joy and simplicity of nourishing yourself.

What it does

Umami is an intelligent culinary co-pilot that bridges the gap between recipe inspiration and actual execution. It solves three critical problems: the fridge-to-recipe gap (not knowing what to make with available ingredients), the saved recipe black hole (hundreds of bookmarked recipes that never get used), and the missing ingredient collapse (abandoning meals mid-cook due to one missing item).

The app features an AI meal planning interface for personalized weekly menus, smart recipe search with filters for cuisine, diet, and dietary restrictions via Spoonacular, and "Pantry Magic" that helps users discover recipes based on ingredients they already have. Users can generate smart grocery lists directly from recipes and save favorites to a persistent library.

How we built it

We built Umami using React Native with Expo Router for file-based navigation, leveraging Reanimated for the Studio Ghibli-inspired gate entrance animation. The tech stack includes:

Spoonacular API for recipe data and complex search with filters for cuisine, diet, intolerances, and prep time Clarifai API integration for the fridge scanning feature (still in development - proper training requires significantly more individuals than our two-person team can currently provide)
AsyncStorage for persistent recipe library management with timestamp metadata
FlashList for optimized recipe card rendering across Pantry, Grocery, and Library tabs

The UI uses custom typography (Roboto for a clean minimilist look) to create a warm, approachable aesthetic that makes meal planning feel less like a chore and more like a comforting experience!

Challenges we ran into

Fridge scanning AI training proved to be our most significant challenge. Food recognition requires massive, diverse datasets and extensive fine-tuning - a task that realistically needs more than two people to execute properly. We've integrated the Clarifai API framework, but achieving production-ready accuracy with real-world fridge photos (messy lighting, expired containers, obscured labels) is still a work in progress that we're planning to tackle with additional resources.

API latency and logic constraints limited our Pantry Magic feature. Spoonacular's response times and recipe variations meant we couldn't always deliver instant, highly personalized suggestions. The external dependency restricted how intelligently we could prioritize recipes based on expiring ingredients or user preference patterns.

State synchronization across tabs proved tricky. Ensuring that a recipe hearted in the Grocery tab immediately appeared in the Library required careful AsyncStorage management and implementing useIsFocused hooks to refresh data when users navigated between screens.

Accomplishments that we're proud of

We're incredibly proud of creating an app that actually solves our own problem. As university students who struggled with consistent meals, seeing Umami eliminate the mental friction of meal planning feels deeply validating.

The Studio Ghibli-inspired design philosophy is something we invested heavily in - those films always made mundane tasks feel magical to us as kids, and we wanted Umami to bring that same warmth to cooking. The gate animation sequence and handwritten UI elements successfully capture that comforting aesthetic.

We're also proud of the infrastructure layer we've built. Unlike typical recipe apps focused on discovery, we own the critical connection between existing ingredients and executable recipes. The real-time filtering system and cross-tab heart logic demonstrate genuine technical sophistication beneath the cozy interface.

What we learned

User empathy drives better design decisions. By building for ourselves first - university students drowning in takeout guilt - we naturally prioritized features that reduce friction rather than add complexity. The "saved recipe black hole" insight came from our own 300+ Instagram bookmarks we'd never cooked from.

AI implementation requires realistic scoping. We learned that some features, like accurate computer vision for food recognition, require resources beyond what a two-person team can deliver in a hackathon timeframe. Training robust AI models demands extensive datasets, computational power, and iterative refinement. Understanding when to mark something as "work in progress" rather than rushing a half-baked feature was an important lesson.

Animation creates emotional connection. The gate entrance sequence took significant iteration, but the investment paid off. Users don't just open an app - they enter a space that feels intentional and welcoming. That emotional resonance matters for habit formation in a meal planning tool.

What's next for Umami

Completing the fridge scanner is our first and top priority. With additional resources that we hope this hackathon would provide, we will be able to properly train the AI for ingredient recognition. This is realistically a multi-person effort that we're excited to tackle post-hackathon!

As mentioned in our business proposal, grocery partnership integrations with Instacart and Amazon Fresh will close the execution loop. One-tap purchasing from generated shopping lists eliminates the final friction point between plan and plate. The revenue projection data is all provided in the document for further understanding.