The product

Stealth Studio

Workflow

Create project

Name a project to start the image selection flow.

Name Images Generate

How it was built

AI collaboration

Codex handled the monorepo setup, Svelte app work, Cloudflare Worker routes, Effect layers, Drizzle schema, deployment wiring, and repeated QA passes.
Specialized agent passes were used for architecture cleanup and product design review while the main implementation stayed in one integrated code path.
I stepped in when generated work drifted from the brief: removing mock outputs, enforcing real Google AI calls, replacing odd Effect dependency passing with tags/layers, and simplifying copied reference-site navigation into utility-only UI.

Time breakdown

Hands-on time ~3h

AI/LLM work ~5h

Approximate focused time, excluding idle periods caused by local machine lag and sleep-prevention stalls.

Toolset

Codex, subagents, and Product Design review: implementation, architecture review, and design critique.
SvelteKit, VitePlus, and TypeScript: web app, local checks, builds, and development server.
Effect and Drizzle: dependency-injected services, repositories, contracts, and typed D1 persistence.
Cloudflare Wrangler, Pages, Workers, D1, R2, and Queues: deployed hosting, storage, database, and generation jobs.
Google AI Studio / Gemini: reference analysis and product image generation with the provided key.
Chrome, local smoke images, Git, GitHub, and CLI checks: visual QA, real generation testing, commits, push, and deployment verification.

The code

Monorepo source for the Svelte web app, Cloudflare Worker server, contract package, database schema, and deployment docs.