“Premature optimization is the root of all evil.”

Donald Knuth
Wednesday, November 18, 2026All the code that's fit to shipLast press run: 3 days ago

Case study, 2026, team of four

XillaFit

Custom sportswear with a live 3D designer, an AI design studio and production tracking.

Visit live site
Visit live site ↗
XillaFit, screenshot.

The stack

Frontend
React, Three.js, Tailwind CSS
Backend
Node.js, Express.js, REST APIs
Database
PostgreSQL, Supabase
Auth
role-based access
Hosting
Vercel
AI
Gemini images, OpenRouter prompts, Background removal

The problem

Ordering custom team apparel usually means long message threads, unclear previews and no view of where an order stands. XillaFit puts the design, the preview, the order and the production steps in one place for customers, staff and the factory floor.

My role

Led the UI/UX and built the client side: the landing page, sign-in, profile and the interface of every module, including how artwork and images are placed on the 3D models. Also wrote some backend functions and reworked the heavy queries. Specialised in the AI studio and did the AI integration: an OpenRouter model that sharpens the customer's prompt, Gemini that generates the design image, and a free background-removal model that cleans up uploaded artwork. Led the performance work when the system struggled under load.

How it works

CLIENTReact, Three.jsTailwind CSSAPINode.js, Express.jsREST APIsDATABASEPostgreSQL, SupabaseAUTHrole-based accessHosted on Vercel
  • A live 3D customizer for team apparel
  • An AI design studio: one model improves the prompt, another generates the image
  • Background removal for uploaded artwork, using a free AI model
  • Checkout, and production tracking for staff

The hard part

Keeping it up under load. I traced slow queries, cut extra realtime subscriptions, and kept real-time working, the right data fetching and every screen responsive.