BeyondCalories.ca
Developed a hackathon recipe platform that indexed 500k+ recipes in Elasticsearch, enabling fast ingredient- and pantry-based search at scale. Combined traditional search, recommendation logic, and vector embeddings to power both exact recipe matching and AI-generated meal suggestions tailored to users’ pantry ingredients and preferences.
BeyondCalories.ca was built during a hackathon to explore how search and recommendation systems could make recipe discovery genuinely useful in everyday life.
The platform indexed roughly 500,000 recipes using Elasticsearch, allowing users to search based on what they actually had in their pantry, along with dietary constraints and personal preferences. The goal wasn’t novelty, it was making recipe discovery practical and actionable.
Core components included:
- Fast, scalable recipe retrieval over a large dataset using Elasticsearch
- Ingredient- and preference-based filtering to quickly narrow down options
- Vector embeddings to enable semantic search beyond exact ingredient matches
- A RAG (Retrieval-Augmented Generation) layer to suggest new or modified recipes based on pantry inputs
- Recommendation logic that surfaced meals that were both relevant and realistic for the user
- A FastAPI backend powering search and recommendations, with a React frontend for interaction
What made the project interesting was the balance between scale and usability. The challenge wasn’t just handling a large corpus of recipes, it was designing a system that could turn that data into meaningful, quick decisions, whether through direct matches or AI-generated suggestions tailored to what users