Veri: AI Vision Analysis-Based Shopping Matching System for Pinterest Fashion Images

Veri is an intelligent search application that turns Pinterest fashion inspiration into purchasable products. Its core uses OpenAI Vision to analyze image style features and combines user preference learning to deliver accurate recommendations. The project addresses the pain point that traditional text searches struggle to accurately convey fashion styles, built with modern tech stacks like Next.js and Supabase, providing a smooth 'import-analyze-match' shopping experience.

For fashion users, Pinterest is a treasure trove of inspiration, but there are difficulties in turning visual inspiration into purchased products: traditional text searches struggle to accurately describe styles (e.g., 'French elegance'), matching involves multiple dimensions like color, style, and fabric, and users' needs for single items or affordable alternatives are unmet. Veri addresses this pain point by combining AI vision analysis with e-commerce search, building a bridge from inspiration to purchase.

Multi-dimensional AI Vision Analysis

• Extract structured information such as category, style, color, and outline based on OpenAI Vision
• Support aggregated analysis of Pinterest boards to generate users' overall style profiles

Intelligent Matching Mechanism

• Multi-modal search (text/image/mixed)
• Semantically relevant query expansion (e.g., expanding 'Bohemian long dress' to 'ethnic-style printed dress')
• Refined filtering (occasion, style, length, etc.)

Interaction Design

• Tinder-style card swiping for browsing: swipe right to like, left to skip
• Preference learning (short-term session interaction + long-term profile + board association)

Technical Architecture

• Frontend: Next.js16 + React19 + TypeScript + Tailwind CSS4
• Backend: Supabase (authentication/data), OpenAI Vision API (image analysis), Playwright + Browserbase (Pinterest scraping)
• Deployment: Vercel optimized configuration

Individual Users

Solve the problem of unbuyable Pinterest collections, meet the 'find similar items' needs for specific scenarios (graduation dresses/vacation outfits)

Fashion Bloggers

Embed product links to monetize content, provide fans with 'one-click get similar items'

Retailers

Get precise traffic, match user style preferences to improve conversion rates

Developers

Provide a complete reference case for AI + e-commerce, covering core links like scraping, analysis, and recommendation

Current Limitations

• Data source dependency: Pinterest scraping is affected by page structure changes; stability challenges in serverless environments
• Product library coverage: matching effect depends on the richness of the database
• Analysis accuracy: details like fabric texture are hard to judge accurately via images

Summary

Veri is an innovative application of AI technology in the e-commerce field, integrating computer vision and traditional search to solve real pain points, providing a learnable case for AI + e-commerce entrepreneurs (pain point positioning + technology combination + experience design)

• Virtual try-on: integrate AR/VR technology to boost purchase confidence
• Social features: friend system and community sharing
• Price tracking: notification for price drops of collected products
• Multi-category expansion: from clothing to visually-driven categories like home goods and beauty products