Chefbooks

Tech Stack

• › React Native, TypeScript, Expo
  Frontend

  Cross-platform mobile development enabling rapid deployment to both iOS and Android with shared codebase.

• › Supabase
  Backend

  PostgreSQL database, Edge Functions, and Storage for serverless backend infrastructure.

• › Apify
  Web Scraping

  Automated content extraction from Instagram, TikTok, YouTube, and recipe websites.

• › Claude Sonnet 3.5
  AI/LLM

  Recipe extraction and structured data parsing via Amazon Bedrock.

• › pgvector with gte-small
  Vector Search

  Semantic similarity search for ingredient-to-emoji matching using embeddings.

Engineering Challenges

• Foundation Model Evaluation

  Tested 9+ different LLMs to figure out which one could actually extract recipes accurately without burning through our budget.

• Real-Time Recipe Import Notifications

  Built a hybrid notification system because waiting 20-60 seconds for AI to process recipes feels like forever without feedback.

• Offline Capability Trade-offs

  Chose to ship fast over building offline-first. Sometimes getting something out there matters more than perfect architecture.

• PostgreSQL Row Level Security

  Used database-level permissions to keep users' recipes private in a multi-tenant setup. Let Postgres do the heavy lifting.

• LLM Prompt Engineering

  Crafted prompts with strict JSON schemas so the AI would consistently parse ingredients instead of getting creative with the format.

• Emoji Ingredient Search Engine

  Built semantic search so typing "tomato" actually finds 🍅 Turns out vector embeddings are pretty good at matching food to emojis.