Inspiration
We've all been there, staring into the fridge with no idea what to cook, or buying groceries only to let half of them go to waste. We wanted to build something that meets you where you are: just tell us what you have, and we'll take care of the rest. InstaPrep was inspired by the idea that great home cooking shouldn't require a perfectly stocked pantry or hours of meal planning.
What it does
InstaPrep turns your available ingredients into personalized recipe suggestions in seconds. Users can scan their fridge with a photo, type out what they have, and our AI identifies the ingredients and generates three tailored recipe options. Each recipe comes with a full ingredient list, step-by-step instructions, prep/cook times, and a food photo. Users can rate recipes, leave feedback, and favourite the ones they love, and InstaPrep learns from that history to make better suggestions every time.
How we built it
Frontend: React + Vite, hosted on Firebase Hosting Backend: FastAPI (Python) serving four endpoints, image analysis, text analysis, recipe generation, and recipe image lookup AI: Groq API with Llama 4 Scout for vision and text ingredient parsing, and Llama 3.3 70B for recipe generation Database & Auth: Firebase Firestore and Firebase Authentication (email/password + Google OAuth) Photos: Unsplash API for on-demand recipe photography
Challenges we ran into
Getting the AI to consistently return structured JSON was trickier than expected, we had to carefully engineer prompts to prevent hallucinated formats and non-food items slipping into the inventory. Merging a user's existing saved inventory with newly scanned items (without duplicates) required some thought around deduplication logic. We also ran into merge conflicts working across multiple teammates, particularly around the recipe generation flow and the inventory review page.
Accomplishments that we're proud of
We're proud that the full loop actually works end-to-end: scan → review → generate → save → influence future suggestions. The personalization layer where liked, favourited, and disliked recipes genuinely shape future AI outputs felt like a meaningful step beyond a basic recipe app. We're also happy with how clean and fast the UI feels given the timeframe.
What we learned
We learned how much prompt engineering matters when relying on LLMs for structured data. We also got hands-on experience building a feedback loop into an AI product not just generating output, but using user behaviour to continuously improve it. On the technical side, the team got more comfortable with Firestore data modelling and async FastAPI patterns.
What's next for InstaPrep Voice-guided cooking ElevenLabs TTS walking users through each step hands-free Dashboard a full inventory management page where users can view and edit their saved ingredient stock Dietary preferences filtering recipes by allergies, dietary restrictions, and health goals set in the user profile Smarter personalization incorporating more signals (cuisine preferences, cooking skill level, time constraints) into recipe generation
Log in or sign up for Devpost to join the conversation.