Inspiration
What it does
🍳 About Recipe Explorer
🌱 Inspiration
I initially explored OpenFood Facts' database for nutritional insights, but realized home cooks need more than raw data—they need actionable recipes. That pivot led me to build a hybrid search engine that bridges food photography with cookable, health-conscious meals.
🛠️ How We Built It
- AI-Powered Search:
- Used Google's Vision AI to convert dish photos into vector embeddings.
- Leveraged MongoDB Atlas Vector Search to match images/text to recipes.
- Used Google's Vision AI to convert dish photos into vector embeddings.
- Health Intelligence:
- Google Gemini analyzes API-sourced recipes, flagging allergens/sugar content (e.g., "Health Score: 3/5 – Reduce sugar by 25%").
- Google Gemini analyzes API-sourced recipes, flagging allergens/sugar content (e.g., "Health Score: 3/5 – Reduce sugar by 25%").
- Guided Cooking:
- Integrated Google Cloud TTS for step-by-step audio instructions.
- Integrated Google Cloud TTS for step-by-step audio instructions.
🧠What I Learned
- Hybrid search isn’t just keywords + vectors—it’s about balancing recall (finding all relevant recipes) and precision (showing only the best matches).
- Small datasets demand creativity: With only 304 recipes, we optimized for quality over quantity (e.g., manual accuracy testing).
🏔️ Challenges
- Data Limitations: Free-tier recipe APIs restricted our dataset. Workaround: Focused on AI accuracy tuning for the available recipes.
- Real-Time Analysis: Gemini’s latency for health insights. Fix: Pre-processed common nutritional flags.
🔮 What’s Next?
- Expand to 10K+ recipes with paid API tiers.
- Add user allergy profiles for personalized filtering.
From OpenFood Facts to a full-stack AI kitchen assistant—this hackathon taught me to pivot with purpose.
Log in or sign up for Devpost to join the conversation.