VibeFit: AI Wardrobe Engine

The Vision

Most fashion apps fail because manual data entry is high-friction. VibeFit solves this by automating wardrobe ingestion. By combining Computer Vision with Vector Search, I built a system that 'understands' a user's style and physique to provide context-aware outfit recommendations.

Screenshots

Previous

Next

• 1
• 2
• 3
• 4
• 5

The AI Ingestion Pipeline

To move away from manual forms, I engineered an automated pipeline using n8n and Gemini 1.5 Flash to extract metadata from raw images.

• Asynchronous image processing via n8n webhooks to prevent UI blocking.
• Computer Vision extraction of attributes (color, pattern, fabric, season).
• Automated JSON schema validation to ensure database integrity.
• Cost-effective scaling using Gemini Flash for high-speed metadata generation.

Semantic Search & RAG

Unlike traditional category filters, VibeFit uses vector embeddings to understand natural language queries like 'what should I wear to a rainy outdoor wedding?'

• Implementation of pgvector for high-performance similarity searches.
• Generating 1536-dimension embeddings for both user queries and clothing items.
• Cosine similarity logic to map user physique data to clothing fit profiles.
• Sub-150ms retrieval time for wardrobe matches from a cold start.

Engineering for Performance

As a Product Engineer, I optimized the stack for a 'premium' feel—instant, resilient, and secure.

• Optimistic UI updates using Tanstack Query for immediate feedback during uploads.
• Edge-runtime deployment for latency-critical API routes.
• Row Level Security (RLS) in Supabase ensuring strict multi-tenant data isolation.
• Zod-validated API contracts to maintain a bulletproof backend.

Technical Trade-offs

• Chose n8n over a custom Node.js worker for visual debugging and faster iteration of AI logic.
• Prioritized pgvector over Pinecone to keep relational user data and vectors in a single ACID-compliant database.
• Implemented a 'Shadow Ingestion' flow where metadata is refined in the background to keep the initial upload fast.

The Outcome

VibeFit is a proof of concept for the future of 'Intent-Based UI.' It demonstrates my ability to bridge complex AI orchestration with a minimalist, high-performance frontend. The result is a production-ready engine capable of handling personalized, multi-modal data at scale.