fridgorithm

fridgorithm: Your AI-powered kitchen sidekick. Snap it, say it, cook it!

Comment

Inspiration

What inspired us to build fridgorithm was, honestly, real life. As students juggling classes, work, and family responsibilities, we know how exhausting it can be to come home and still have to figure out what’s for dinner. We’ve all been there — scrolling through recipes, only to realize we’re missing that one key ingredient. Sometimes, after a long day, you just want something quick and easy without turning dinner into another research project.

That’s where the idea for fridgorithm was born — an app that takes the mental load off. Whether you snap a photo, speak your ingredients, or type them in, the AI does the thinking. No stress, no guesswork — just pass that homework to the AI and get cooking.

What it does

fridgorithm is a web app that takes the ingredients you already have — whether you snap a photo, say them out loud, or type them in — and instantly gives you a simple, delicious recipe. It’s designed to save time, reduce food waste, and take the guesswork out of cooking. Powered by AI, fridgorithm turns your fridge contents into ready-to-make meals in just seconds.

How we built it

We built Fridgorithm with a React frontend that connects to an Express.js backend via API calls. On the AI side, we trained a Custom Vision model using Azure, running five iterations with seven distinct ingredient tags. After fine-tuning, we achieved a 72% accuracy rate in identifying common fridge items — a strong starting point for a lightweight, fast prototype.

We also integrated Azure Speech-to-Text to power our voice input feature, allowing users to speak their ingredients hands-free. Regardless of input method — whether typed, spoken, or image-based — all data is ultimately converted into structured string format and sent to the Azure OpenAI GPT-4.0 Turbo model for recipe generation. We provided the model with carefully crafted custom system instructions with specific roles, guiding it to interpret input as a list of ingredients and respond with concise, easy-to-follow recipe suggestions. This seamless pipeline between input processing and AI response is what allows Fridgorithm to deliver fast, context-aware results in seconds.

Challenges we ran into

As third-semester students at Sheridan College, this was only our second hackathon — and we really saw it as a chance to challenge ourselves and grow. One of the biggest learning curves was React. It’s not part of our curriculum, so diving into a brand-new framework while trying to apply our foundational knowledge of single-page applications was definitely a stretch.

We also ran into challenges making the app responsive on mobile browsers. Ensuring a smooth, functional experience across devices took a lot of trial and error, and there are still areas we’d like to improve.

The steepest learning curve, though, was integrating and training AI — something we hadn’t touched in our coursework. We had to figure out how to tag images properly, use techniques like negative tagging, and decide what kinds of photos were best for training. Our first few iterations were honestly a bit discouraging, and we weren’t sure if we could get the model to a usable level. But…

Accomplishments that we're proud of

Despite all the hurdles, we’re incredibly proud of what we accomplished. Learning React on the fly and managing to build a clean, responsive single-page app was a huge win for us. We’re also proud of how we took on something we’d never done before — training a Custom Vision model completely from scratch — and getting it to a point where it actually worked well in practice.

We’re also proud of how we pulled everything together: voice input, image recognition, AI-generated recipes — all working seamlessly in one user-friendly app. For us, it wasn’t just about building something functional, but creating something meaningful that solves a real problem. Fridgorithm started as a learning project, and turned into something we’re truly excited to keep building.

Just as important as the tech, we’re proud of how we worked together. Everyone on the team took initiative, carving out time whenever we could — between classes, assignments, and life — to push this project forward. We supported each other, filled in gaps when someone got stuck, and built each part of fridgorithm with care and collaboration. It really felt like a team effort from start to finish.

What we learned

This hackathon pushed us outside of our comfort zones, and we learned a lot. We gained hands-on experience building a full single-page application with React, and learned how to connect it to an Express.js backend and manage API calls efficiently.

Working with Azure’s AI tools was another huge learning opportunity. We explored how to train a Custom Vision model, and picked up practical strategies like tagging best practices, negative tagging, and improving model accuracy through iteration. We also learned how to integrate Azure’s Speech AI and OpenAI services, and how to structure those workflows on the backend.

Beyond the tech, we also learned how to collaborate more effectively under pressure, divide tasks based on our strengths, and adapt quickly when things didn’t go as planned. Most importantly, we learned that we’re capable of a lot more than we thought when we give ourselves permission to try something new.

What's next for fridgorithm

We’ve got a lot of exciting ideas in the pipeline to take fridgorithm even further:

  • Multiple Recipe Options: Instead of just one result, we plan to generate several recipe suggestions and let users toggle between cards to choose the one they’d like to view in detail.
  • Smarter Ingredient Recognition: We’re training our vision model to recognize way more ingredients — not just the basics, but diverse and culturally relevant ones, with better accuracy and broader training data.
  • Personalized Experiences: We want to give users the ability to save personal preferences, such as dietary restrictions, favourite cuisines, or cooking skill level, to receive more tailored recipe suggestions.
  • One-Click Grocery Shopping: Down the line, we hope to integrate with grocery delivery apps so users can instantly order any missing ingredients straight from their recipe cards.

And so much more! This is just the beginning. We’re excited to keep iterating, learning, and building features that make fridgorithm smarter, faster, and even more helpful in the kitchen.

Built With

Share this project:

Updates