Fit Homie

[ Overview ]

Fit Homie is a LINE-first fitness and nutrition agent that helps users track meals, log workouts, and get micronutrient-level health insights through LINE Messenger. It uses Gemini AI to extract nutrition from food photos — snap a meal, get a full nutritional breakdown including 24+ vitamins and minerals. The architecture features a 5-layer cost optimization pipeline to keep AI inference affordable at scale, a multi-layer nutrition cache for frequently logged foods, and tiered subscription billing. Built after the user asked "my ex girlfriend and friends got hooked" on early versions in Thailand.

[ Key challenges ]

• • AI cost management at scale — 5-layer cost optimization pipeline to keep Gemini API calls affordable while processing meal photos
• • Food recognition accuracy across Thai cuisine — regional dishes, mixed plates, and ambiguous presentation
• • Multi-layer nutrition caching — deduplicating frequent foods while handling recipe variations and portion differences
• • LINE Messaging API limitations — rich menu constraints, message size limits, and platform-specific UI paradigms
• • Micronutrient database — mapping recognized foods to 24+ vitamins and minerals with daily value percentages

[ Key features ]

• • AI food extraction from photos — snap a meal, get instant nutritional breakdown with 24+ micronutrients
• • LINE-first architecture with rich messaging, interactive flex menus, and quick reply buttons
• • 5-layer cost optimization pipeline — image preprocessing, cached recognition, partial fallback, tiered model selection, result caching
• • Multi-layer nutrition cache reducing API costs by 60%+ for frequently logged foods and restaurant meals
• • Tiered subscription billing with Stripe — free tier, $9.99/mo premium with full AI and micronutrient analysis
• • Smart nutrition alerts — real-time deficiency and excess flags with context and actionable suggestions
• • AI coach chat for fitness and nutrition advice with personalized recommendations based on logged data

[ Screenshots ]

[ Tech stack ]

• • Django backend with LINE Messaging API integration for rich messaging and interactive menus
• • Gemini AI for food recognition, nutritional extraction from photos, and natural language coaching
• • Next.js dashboard for nutrition analytics, progress visualization, and account management
• • 5-layer cost optimization pipeline — preprocessing → cached match → partial → full inference → result caching
• • Stripe for tiered subscription billing with webhook-driven access management
• • PostgreSQL for user profiles, meal logs, workout history, and nutrition cache
• • Redis for real-time rate limiting, session management, and hot cache for frequent food lookups