Tastory

Tastory is a recipe search engine built on top of Food.com public dataset. You can search from over half a million recipes, delightful homemade food and cocktails.

Comment

Inspiration

Every culture has stories woven into its food. We wanted to create a way to discover global recipes not just by name or ingredient, but by feeling. Whether you're nostalgic, adventurous, or homesick—Tastory helps you find the dish that matches your current vibe. We built it with the idea that food is memory, mood, and meaning, all mixed in a pot.

What it does

Tastory is a recipe discovery platform powered by vector search and AI. Users can:

  • Search for recipes based on ingredients or free-text descriptions
  • Retrieve globally inspired dishes from our MongoDB collection
  • View results with rich metadata including description, cuisine, ingredients, and AI-suggested image
  • Explore culturally rooted dishes that align with personal moods or cravings

How we built it

  • Frontend: React with Tailwind for a clean, responsive UI
  • Backend: Python (Flask) handling the API endpoints
  • Database: MongoDB Atlas with vector search enabled using embeddings
  • Embeddings: Initially used Sentence Transformers; transitioned to Google Vertex AI Embedding API
  • Image Augmentation: Used food.com dataset; supplemented missing images via external datasets/models

Challenges we ran into

  • Dataset inconsistencies: Several entries lacked proper structure or images
  • Embedding limitations: Had to transition from sentence transformers to Google Vertex AI to stay within the Google ecosystem
  • Latency: Optimizing vector search queries for performance
  • Deployment hurdles: Learning curve around GCP, setting up Cloud Run & enabling necessary APIs

Accomplishments that we're proud of

  • Successfully deployed an AI-powered app on Google Cloud
  • Enabled semantic search over recipes using real embeddings
  • Overcame missing data issues and still built a visually compelling experience
  • Made food discovery emotional and global, not just logical

What we learned

  • How to set up and scale MongoDB Atlas with vector search
  • How to integrate Google Vertex AI for custom embedding generation
  • Real-world problems of multilingual, multicultural data
  • How to tell a story with technology, not just serve a function

What's next for Tastory

  • Add user accounts to save favorite recipes and moods
  • Enable voice-based search for accessibility and natural interaction
  • Partner with culinary creators to expand and validate the dataset
  • Use LLMs to dynamically generate new fusion recipes
  • Add cultural notes and history blurbs to enrich each dish

Built With

Share this project:

Updates