JPT Logo
Jose Paulo Timbang
All projects

pamangan.com

Live

AI-powered Filipino recipe platform

Tech Stack

React 18FlaskPythonMongoDB AtlasGemini 1.5 FlashGroqPyMongoBootstrap 5KoyebHostinger

Overview

pamangan.com is a full-stack AI-powered recipe platform celebrating Filipino cuisine and global flavors. Users can discover recipes, generate new ones via AI, build grocery lists from multiple recipes, get per-serving nutrition estimates, and plan meals for the week. The platform uses a database-first search strategy to minimise unnecessary API calls.

Key Features

  • AI recipe generation using Gemini 1.5 Flash (primary) and Groq llama-3.3-70b (fallback)
  • Database-first search — MongoDB text index checked before any AI API call
  • AI grocery list generation from multiple selected recipes
  • AI nutritional breakdown per serving with macro estimates
  • Cultural and historical background for any dish
  • Weekly meal planner saved to localStorage
  • Admin dashboard with JWT authentication and recipe CRUD
  • Pexels image integration and ImgBB upload for recipe photos

Challenge

Coordinating two AI providers with silent fallback so users never see an error. Also ensuring the database-first search was fast enough to avoid unnecessary API calls on every keystroke.

Solution

Built a service layer that tries Gemini, catches any failure silently, and immediately retries with Groq. MongoDB text indexes on recipe name, cuisine, and tags made DB lookups fast enough to always run first.

Lessons Learned

AI reliability is not guaranteed — always design for graceful fallback. MongoDB text indexes are powerful but require careful field selection. Separating AI logic into a service class makes it trivially swappable.

P

Paulo's AI Assistant

Online