VisionFit AI
AI-powered fashion recommendation platform that uses Gemini AI to provide personalized style suggestions based on user preferences, body measurements, and current fashion trends from major retailers like Zara and H&M. Built with a Turborepo monorepo architecture featuring Expo mobile app and Next.js backend.
Problem
Style-conscious individuals struggle to discover apparel and accessories that complement their unique body composition, aesthetic preferences, and occasion-specific requirements. Traditional shopping experiences lack intelligent curation and require hours of manual consultation and browsing through non-relevant options.
- Manual styling consultations take hours and lack scalability
- Difficulty discovering apparel that complements individual body composition and proportions
- Lack of intelligent curation based on personal aesthetic and lifestyle preferences
- Challenges in selecting appropriate attire for specific occasions and social contexts
- Time-consuming process of browsing through non-relevant wardrobe options across multiple retailers
- Limited access to real-time product availability from premium retail sources
Solution
A Turborepo monorepo with a mobile app built on Expo/React Native and a Next.js backend with Hono API routes. The system uses OpenAI vision models for body composition analysis and Gemini AI to generate outfit recommendations by querying pre-scraped product data from fashion retailers stored in PostgreSQL.
- Developed React Native mobile application with OpenAI vision models for body composition analysis, cutting consultation time from hours to minutes
- Built automated web scraping pipeline with Puppeteer significantly reducing infrastructure costs through intelligent caching and batch processing
- Implemented Stripe payment integration and real-time recommendation engine for seamless user experience
- Created Python-based web scraping service that collects product data from Zara and H&M
- Implemented data pipeline: scraping → mapping to canonical schema → database ingestion via Drizzle ORM
- Designed AI tool system with database query tools for product search, filtering, and personalized recommendations
- Implemented two-phase AI generation: product search followed by structured outfit composition
- Wrote comprehensive unit tests with Pytest for API endpoints and Jest for UI components
Outcome
The platform successfully delivers personalized wardrobe curation that aligns with users' body composition, aesthetic preferences, and occasion requirements, significantly reducing consultation time and enhancing the discovery experience.
- Reduced styling consultation time from hours to minutes using AI-powered analysis
- Significantly reduced infrastructure costs through intelligent caching and batch processing
- Accurate body composition analysis and personalized wardrobe curation with OpenAI vision models
- Context-aware suggestions for different occasions and social contexts
- Seamless integration with Zara and H&M product catalogs via automated scraping
- Fast AI responses by querying pre-scraped data from PostgreSQL database
- Scalable data updates via scheduled cron jobs for weekly product refresh
- Improved user satisfaction with relevant and personalized style choices
Challenges
Building a hybrid architecture that separates data collection from API serving while ensuring fast AI responses, maintaining fresh product catalogs from multiple retailers, and implementing accurate body composition analysis required careful system design.
- Designing reliable web scraping with Puppeteer that adapts to changing Zara and H&M website structures
- Creating a canonical data schema that normalizes products from different retailer formats
- Implementing efficient database ingestion with proper normalization across multiple tables
- Building AI tool system that enables Gemini to query products effectively
- Optimizing infrastructure costs while maintaining data freshness and performance
- Ensuring accurate body composition analysis with OpenAI vision models
- Ensuring consistent Turborepo builds across backend, mobile, and shared packages
Key Learnings
Gained expertise in building AI-powered mobile applications with Turborepo monorepos, integrating multiple AI models (OpenAI and Gemini), creating automated data pipelines, and optimizing infrastructure costs through intelligent caching strategies.
- Mastered OpenAI vision model integration for body composition analysis
- Developed expertise in automated web scraping with Puppeteer and cost optimization strategies
- Learned to build scalable backends with Hono and Drizzle ORM in a Turborepo monorepo
- Gained expertise in designing AI tool systems for database-querying assistants
- Mastered cross-platform mobile development with Expo, React Native, and TypeScript
- Developed comprehensive testing strategies with Pytest and Jest
- Learned to integrate Stripe payment processing in mobile applications