Inspiration

FridgeSnitch – A Visual Extension for Smarter Meal Planning

FridgeSnitch is a visual ingredient recognition tool built as a natural extension of the MealestroAI platform — an AI-powered assistant designed to help busy families eat healthier, reduce food waste, and simplify meal planning.

Inspiration

MealestroAI is a project I started last year. I presented this project at one of the AI Tinkerers meetups in Ottawa in October 2024.

The inspiration came from a common pain point: opening the fridge, staring at a mix of half-used ingredients, and still wondering “What should I make for dinner?” Despite building MealestroAI to generate personalized weekly meal plans, I realized the tool needed a more convenient way for users to communicate what they already had at home.

Typing out a list of ingredients is tedious — snapping a photo is not.

What I Learned

This project deepened my understanding of combining multimodal AI (vision + text) into real-world workflows. I explored:

How image analysis with models like GPT-4o can extract high-level semantic meaning (e.g., recognizing “cucumber” vs “green vegetable”).
The importance of user experience in low-confidence detection, feedback loops, and editability of AI-generated results.
I was positively surprised by the quality of GPT-4o’s recognition based on my own fridge images. I’m aware, however, that this will require much more testing.

How I Built It

The frontend is a simple Next.js application. Unfortunately, I failed to deploy it to Vercel — only a short video is provided. However, I have a fully working instance on my local laptop, and I’ve attached a link to the public GitLab repository for free access.
Image inputs are passed to OpenAI’s GPT-4o, which analyzes the photo and returns a structured list of detected ingredients.
The application has been built using Cursor AI.
A feedback layer allows users to review, confirm, or edit the list before sending it to MealestroAI.
The output is integrated with the MealestroAI dinner planning engine, so detected ingredients can be reused in upcoming dinners — saving money and minimizing waste.

Challenges I Faced

Balancing automation and control: Fully automating ingredient recognition isn’t reliable. I needed a lightweight UX that allowed for human-in-the-loop corrections without feeling like extra work.
Ingredient quantity is not yet supported: The current implementation provides only the names of ingredients, not the quantities. A more advanced model may be required in the future to support inventory tracking.
Seamless integration with MealestroAI: The ingredient list needed to feel like a natural part of the planning flow, not a separate process. Stored ingredients will now be taken into account in the next meal plan, and users will be prompted to provide fridge content two days before the dinner plan is generated.

What it does

This app allows users to provide images of the fridge or other places where ingredients are stored and analyzes those images and extract ingredients that can be used in dinner plan for upcoming week. This way it allows to reduce food waste and save money spent on food.