Spread 03
The stack, and why.
Next.js 15 and React 19, TypeScript end to end. One language for the page, the API and the database layer means one small team owns all of it. Server components render product pages on the server and ship almost no JavaScript where none is needed — while the cart, the configurator and the chat stay fully interactive where it counts.
tRPC 11 instead of REST or GraphQL: sixteen routers — products, cart, orders, multisplit, financing, SEO — typed from the database to the component, with no schema to maintain and no client to generate. The same procedures run at build time to pre-render pages and at runtime to hydrate the browser, so the storefront never fetches the same data twice.
PostgreSQL on Neon, Drizzle ORM. Thirty-six tables, typed SQL, versioned migrations. Postgres is boring in exactly the way a shop's database should be: any developer can inherit it. It also carries the assistant's memory and vector index through pgvector, so there is one database to back up, not three.
Better Auth for accounts (e-mail and Google), Tailwind CSS 4 with shadcn/Radix primitives for an accessible design system in weeks rather than months, Resend + React Email for seven transactional templates, and Zod on every input and every environment variable — the build fails if a secret is missing, which is cheaper than finding out in production.
Bun, Vercel and GitLab CI. Bun for fast installs and builds; Vercel for the edge CDN, incremental static regeneration, cron jobs and analytics; GitLab CI for the gates — lint, typecheck and tests run in parallel on every push, main deploys to a test environment hidden behind admin login and marked no-index, and production is a deliberate, manual step. Boring where boring is right, sharp where it pays.