3D Prints Design
3D Prints Design is a small 3D-printing business, and this is its storefront. Customers pick a kit, type a name, and choose two filament colors. The site generates a print-ready model on demand from a parametric template, shows it in an interactive 3D preview, and takes the order.
- My role
- Team of three: I wrote most of the code, across the storefront, the API, the model generator, and the deployment pipeline
- Topics
- Next.js · TypeScript · Fastify · OpenSCAD · PostgreSQL · Stripe · Kubernetes
The Idea
Personalized prints are a natural fit for 3D printing, but designing each model by hand doesn't scale. The idea was to make the customer's input the design: a name and two colors go in, and a ready-to-print model comes out, with no designer in the loop.
What I Built
A storefront with a product studio, cart, checkout, and order and admin pages, backed by an API that generates every model on demand. Each order produces the printable parts and a two-color 3MF file ready for the printer, and the same files drive the live 3D preview the customer rotates before buying.
How It Works
It's a TypeScript monorepo: a Next.js storefront, a Fastify API, a shared package that defines the typed contract between them, and a generation engine that drives the OpenSCAD command-line tool from a parametric template. The API runs generation jobs through a queue, the browser polls for the result, and checkout goes through Stripe with prices calculated on the server rather than trusted from the browser. Orders are stored in PostgreSQL.
Every merge runs CI (type checks, the build, end-to-end tests against mocked services, and Helm chart validation), publishes container images, and cuts a versioned release. Releases deploy to a small Kubernetes (K3s) cluster with Helm, pinned by image digest, with health checks, zero-downtime rollouts, and automatic rollback if the new version isn't healthy.
Challenges & Decisions
OpenSCAD is a desktop CAD program, not a web service. Running it on a server meant packaging the command-line tool inside the API's container, plus a virtual display, only because its PNG preview export needs OpenGL; the model exports themselves are fully headless. Generation is also CPU-heavy, so jobs run one at a time through a queue instead of competing for the machine.
The other half was making releases safe for a real store: moving from Docker Compose to Kubernetes, testing failure cases during the migration, and setting up a release process that rolls itself back instead of relying on someone to notice a problem.
What I Learned
Owning the whole path, from a customer's keystroke to a file on the printer, shows where the real work is. It isn't the storefront; it's the generation pipeline, the payments, and a deployment process you can trust when it's someone else's order.
Current State
Live at 3dprintsdesign.com.
