DoneDish - See it, Cook it, Done!

• 

  All-in-One Recipes Inventory

• 

  Manage Ingredients at one place

• 

  Snap it and record your ingredients

• 

  Get notified with Expiring Ingredients

• 

• 

  Import Recipes

• 

  Step by Step with Hands-Free Mode

• 

  Grocery List

• 

  Built Around You. Personalisation

Inspiration

The idea for DoneDish stems from a frustratingly common pattern: we discover amazing recipe videos online, save them, share them, and then... nothing happens.

What stops us isn't a lack of motivation; it's friction.

• Choice Overload: Too many recipes, not enough clarity.
• Mental Load: The exhausting effort of manual planning.
• Inventory Blindness: Uncertainty about what’s already in the fridge.
• Waste: Duplicate purchases and forgotten, expired ingredients.

Inspiration turns into overwhelm, and overwhelm turns into inaction. DoneDish was built to bridge the gap between "This looks good" and "Dinner is served."

---

What It Does

DoneDish supports the entire cooking journey—from ingredient awareness to final execution.

1. Know What You Have (AI-Powered Tracking)

• The Process: Users take a photo of their fridge or pantry.
• The Tech: On-device computer vision performs image segmentation and classification to identify items.
• The Benefit: Eliminates manual data entry, prevents food waste via expiry notifications, and forms the foundation for recipe suggestions.

2. Decide What You Can Cook (Inventory-Aware Recommendations)

• The Process: The app analyzes your current inventory, available hardware, and dietary preferences.
• The Tech: Recipes are scored based on "Readiness" (percentage of ingredients owned), difficulty, and prep time.
• The Benefit: Reduces choice paralysis and ensures you only see recipes you can actually finish.

3. Adapt Recipes to Reality (AI Substitution)

• The Process: If you're missing an ingredient, the AI "chef" layer modifies the recipe.
• The Tech: Uses semantic similarity to suggest substitutions and dynamically rewrites cooking steps to accommodate changes.
• The Benefit: Removes dependency on exact ingredients and saves unnecessary trips to the store.

4. Turn Inspiration into Action (Automated Grocery Lists)

• The Process: Input a YouTube or website link.
• The Tech: AI parses unstructured content to extract ingredients and quantities, then "diffs" them against your current pantry.
• The Benefit: You only buy exactly what you need.

5. Cook Without Hassle (Hands-Free Voice Guide)

• The Process: Navigate recipes using voice commands.
• The Tech: Uses Text-to-Speech with natural pausing and (planned) voice recognition.
• The Benefit: Keeps screens clean and prevents cross-contamination.

---

How We Built It

We utilized a Local-First Architecture to ensure reliability in kitchens with spotty Wi-Fi.

Component	Technology
Framework	React Native (Expo SDK 54)
Database	WatermelonDB (SQLite) for offline-first speed
Sync/Auth	Supabase
AI Brain	Google Gemini 2.0 Flash (Vision, Text, Logic)

Key Feature Implementations

• Smart Inventory: Integrated react-native-vision-camera with Gemini 1.5 Flash to return structured JSON arrays of food items.
• Hybrid Ranking: A local algorithm calculates "High Viability" meals in real-time.
• Recipe Tailoring: We swap static recipe views with dynamic "Tailored Recipes" generated by the LLM based on user-specific constraints.

---

Challenges We Ran Into

1. Cost Management: AI infrastructure and API usage can be prohibitively expensive. We mitigated this by moving toward edge models and local caching.
2. Performance Trade-offs: Balancing the speed of image recognition with accuracy while running on-device.
3. Market Differentiation: Designing a unique, intuitive UX to stand out from generic recipe apps.

---

Accomplishments We're Proud Of

• End-to-End Coverage: We support the entire workflow, not just one piece of it.
• Offline AI: Recipes and tailored substitutions are cached locally; you can cook in a basement without a signal.
• Edge Efficiency: Used Gemini 2.0 Flash Lite to keep inference costs low while maintaining high intelligence.
• High Performance: Achieved 60fps scrolling and sub-16ms search times using FlashList and WatermelonDB.

---

What We Learned

• Beyond Giant Models: One-size-fits-all LLMs are over. Task-specific, smaller models are often more effective.
• Orchestration Over Size: How you connect models matters more than the raw parameter count.
• Prompt Engineering: Thoughtful tuning and domain-specific knowledge bases (RAG) can make small models punch way above their weight class.

---

What's Next for DoneDish

• Global Scaling: Targeted marketing and strategic partnerships.
• Cultural Personalization: Adapting recommendations based on geography and cultural preferences.
• Grocery Integration: Seamless API connections with Grab, Uber Groceries, and more.
• Shared Inventories: Syncing the fridge across all household members.
• Health Intelligence: Surfacing nutrition scores and advocating for healthier dietary habits.

Built With

• expo.io
• gemini
• react-native
• supabase
• text
• tflite
• watermelondb