My Stack

This is the shape my side projects keep settling into. Most of it comes from building Fisle in the evenings — there's more context on that project in this post. The constraints (low cost, small team, mostly serverless) pushed me toward a specific stack and away from a lot of things I used to default to.

I write everything in TypeScript.

Frontend

The user-facing app and the marketing site live in the same Next.js project, deployed on Vercel.

• Next.js 16 + React 19 — App Router, Server Components by default. I drop into client components only where I actually need state or browser APIs.
• Tailwind v4 — utility-first. With Radix Primitives I don't reach for a full component library anymore.
• Radix UI — unstyled accessible primitives (dialog, dropdown, popover, etc.) wrapped with Tailwind.
• TanStack Query — for any client-side fetching that has a meaningful lifecycle (mutations, optimistic updates).
• Zustand for small client state, react-hook-form + Zod for forms.
• Motion (formerly Framer Motion) for transitions.
• Serwist for the PWA / service worker layer.
• MDX for content (blog posts, guides). Frontmatter parsed with gray-matter.

Backend

The API runs as a Cloudflare Worker. Same TypeScript, much smaller surface.

• Hono — tiny, fast router that feels right at home on Workers.
• Drizzle ORM — typed schema, no codegen, plays well with D1.
• Zod at the boundaries for input validation.
• Vercel AI SDK for talking to the model providers.
• Sharp for image preprocessing before things go to vision models.
• Wrangler for local dev and deploys. Cron triggers handle the daily housekeeping jobs.

Data

• Cloudflare D1 — the primary app database. SQLite at the edge, cheap, low-friction, fast enough for everything we do.
• Cloudflare R2 — object storage for receipts and uploads. S3-compatible API but without egress fees, which matters when users re-download their files.
• PostgreSQL for a couple of additional internal services.

AI

We use a mix of providers in different parts of the pipeline.

• OpenAI
• Azure (Document Intelligence)
• Google Gemini

Different models are good at different things, and the cost/latency profile changes depending on what part of the flow you're in. So we don't pick one and live with it — we route based on the task.

Auth

• Better Auth in both the Next.js app and the Workers API. Single source of truth for sessions, easier to keep client and server in sync than rolling something custom.

Hosting

• Vercel for the Next.js app.
• Cloudflare for everything else — DNS, Workers, D1, R2, and the proxy/cache layer.

Why this shape

Most of the choices come from one constraint: a side project has to be cheap to run when nobody is using it, and not break when a lot of people are. Cloudflare's free tier covers the API, the database, and the storage for the entire current load. Vercel covers the frontend. The AI providers are the only line that scales with usage, and that's by design.

The other thing is that I want to keep the surface small enough that one person can hold it in their head. Hono + Drizzle + D1 is a much smaller mental model than Express + Postgres + Redis + a queue worker. When something breaks at 11pm, I want to be able to fix it without paging in a microservices diagram.

Most of my recent writing about this stack lives over at fisle.co/blog — practical notes from actually shipping with it.