Matthew Raynor Photography Store
Full e-commerce platform for fine art photography with an AI shopping assistant, semantic search, and wall visualization.
Project Overview
A complete e-commerce platform for fine art drone and seascape photography targeting the Hamptons luxury art market. Features an AI shopping assistant built with LangChain and Claude that can search photos semantically, manage carts, visualize prints on customer walls using depth estimation, and handle checkout through conversation.
The Challenge
Selling fine art photography online requires more than a product grid — customers need to discover art by mood and meaning, visualize how pieces look in their space, and feel confident about size and materials before purchasing.
The Solution
Built an AI shopping assistant with 14 tools that searches photos semantically using pgvector embeddings, manages carts, filters by color/mood/subject, checks gift card balances, and answers sizing questions. The 'See It In Room' feature uses MiDaS depth estimation + RANSAC plane-fitting to composite prints at correct scale on customer-uploaded wall photos.
Technology Stack
Backend
Frontend
Ai
Integrations
Deployment
Key Features
AI shopping assistant with 14 tools — search, cart, checkout, wall visualization all through conversation
pgvector semantic search using OpenAI text-embedding-ada-002 for meaning-based photo discovery
Claude Vision auto-generates all photo metadata (descriptions, moods, colors, subjects)
'See It In Room': MiDaS depth estimation + RANSAC plane-fitting composites prints at correct scale on wall photos
Stripe Checkout with gift card redemption and promotional codes
Session-based cart persistence with cross-origin cookie handling
Next.js App Router with server-side rendering — server components (internal API) vs client components (public API)
Business Impact
Semantic photo discovery — customers find art by meaning, not just keywords
Realistic wall visualization reduces purchase hesitation for expensive prints
Conversational commerce handles the entire shopping experience through the AI assistant
Automated metadata generation eliminates manual photo tagging
Technical Achievements
MiDaS + RANSAC pipeline that accurately places prints on real walls
Semantic search that understands 'moody ocean sunset' or 'bright aerial beach'
Full conversational commerce — customers can browse, add to cart, and check out without leaving the chat
Claude Vision metadata pipeline that auto-tags every photo
Future Enhancements
AR-based room visualization using device camera
Multi-currency support for international buyers
Artist collaboration marketplace
Technical Implementation
Photo embeddings generated with OpenAI text-embedding-ada-002, stored in PostgreSQL with pgvector for cosine similarity search. The AI assistant uses LangChain with Claude and 14 tools for a complete shopping experience. The 'See It In Room' feature uses MiDaS depth estimation to find walls in uploaded photos, then RANSAC plane-fitting to composite prints at physically accurate scale. Next.js App Router separates server components (which call internal API) from client components (which use the public API).