Weir: AI inference website design

Weir is a concept marketing site and docs for an invented AI-inference company in San Francisco. It serves open-weight models like Llama, Qwen and DeepSeek on shared GPUs, billed per token, and on dedicated GPUs, billed by the hour.

The brief: inference companies have a hard site to build. Dozens of models, prices that change with the model and the type of capacity, and enterprise buyers who want SLAs and compliance answers. The site had two jobs: let a developer find a model and price it in a minute, and give an enterprise buyer what they need before the first sales call.

What I built

• A model catalog with task filters, price sorting and animated rows

• A command menu (Ctrl K) across pages and models

• A side-by-side model comparison with both monthly bills, shareable by URL

• A pricing calculator: serverless per token vs. one dedicated GPU, with the break-even volume

• An enterprise page with an accessible, validated contact form

• Quickstart docs written in MDX

How it was built

• Next.js 16 (App Router, static export), React 19, TypeScript, Tailwind CSS v4, shadcn/ui on Radix and Motion

• Every price and benchmark comes from one data source, and a test checks that they match on every page

• PageSpeed 98 to 100 on mobile and 100 on desktop, with accessibility, best practices and SEO at 100

• Vitest and Playwright tests, axe accessibility checks and CI on every pull request

• A Figma design system whose variables mirror the code tokens

• Built with Claude Code as an AI pair programmer, in small pull requests

Live site: https://weir-inference.netlify.app

Code: https://github.com/LFUrbano/weir