Reciprep: Stop Scrolling, Start Cooking

Reciprep bridges the gap between scrolling and cooking. With recipe extraction, smart grocery & pantry, and persistent reminders, we ensure your saved recipes don't just sit there, they become dinner.

Comment

Inspiration

It started with late-night scrolling. We realised we were saving hundreds of recipe videos on TikTok and Instagram, perfect, viral, delicious-looking meals, but never actually cooking them. We call this the Inspiration Gap.

The gap between seeing a video and eating the meal is filled with friction: transcribing ingredients, checking the pantry, making a list, and remembering to cook before the produce goes bad. We built Reciprep to close this loop. We wanted an app that didn't just bookmark content but actively pushed us to cook it.

What it does

Reciprep is an AI-powered kitchen assistant that transforms social media videos into actionable cooking plans.

  1. AI Extraction: Transcribes video/audio to structured recipes.
  2. Ingredient Intelligence: Cross-references recipes with your digital pantry.
  3. Persistent Intent: Smart reminders that nudge you until you cook.

How we built it

  • Frontend: Flutter (Dart) for a beautiful, responsive mobile UI.
  • Backend: Python (Flask) for robust API handling and business logic.
  • AI: Google Gemini 3 Pro for multimodal extraction (Video $\to$ Text) and ingredient reasoning.
  • Database: SQLAlchemy (Relational DB) to model the complex relationships between Users, Recipes, and Pantry Items.
  • Notifications: Firebase Cloud Messaging (FCM) for our "Persistent Intent" system.

The Algorithm

To determine if a user can cook a recipe $R$ given their pantry $P$, we calculate a Readiness Score $S$:

$$ S(R, P) = \frac{|R_{ingredients} \cap P_{items}|}{|R_{ingredients}|} $$

If $S < 1$, the missing set $M = R_{ingredients} - P_{items}$ is automatically added to the user's shopping list.

Challenges we ran into

  • Video Transcription Accuracy: Short-form content is fast and chaotic. Creators often skip mentioning quantities. We had to prompt engineering Gemini to "infer" reasonable amounts based on visual cues (e.g., "a handful of spinach" $\approx$ "1 cup").
  • Ingredient Hallucinating: Early models would invent ingredients or URLs. We implemented strict validation layers to ensure only real, non-hallucinated data made it to the database.
  • State Synchronisation: Keeping the pantry state in sync between the "Home" dashboard and the "Recipe Detail" view required complex state management with GetX controllers, especially when toggling ingredients on/off in real-time.

Accomplishments that we're proud of

  • Seamless Extraction: The "Paste Link $\to$ Full Recipe" flow feels magical. It takes a raw URL and returns a structured object with prep time, difficulty, and nutritional estimates in seconds.
  • Real-time Pantry Logic: Seeing the ingredient badges flip from "Missing" (Red) to "In Pantry" (Green) instantly when you update your stock feels incredibly satisfying and useful.
  • The "Clean Up" Protocol: We built a logic that automatically deducts used ingredients from your pantry after you mark a recipe as "Cooked," keeping your inventory accurate without manual work.

What we learned

We learned that friction is the enemy of cooking. Even a small barrier, like not knowing if you have garlic, stops people from trying a new recipe. By using AI to remove that friction, we can actually change user behaviour from passive consumption to active creation.

What's next for Reciprep: Stop Scrolling, Start Cooking

  • Social Cooking Challenges: Challenge friends to cook the same "viral" recipe and share results.
  • Smart Home Integration: Sync with IoT fridges to automatically update the digital pantry.
  • AR Cooking Assistance: An augmented reality overlay that shows steps and timers directly on your countertop.
  • Meal Prep Mode: Aggregate ingredients from multiple recipes into a single efficient prep session.

Built With

Share this project:

Updates