Ember AI

I developed Ember, an AI-powered web and app builder that turns a plain-English description into a running, deployable product. It pairs a streaming chat interface with a live preview and full Git history, so users can describe an idea, watch the code being written, iterate through follow-up prompts, and ship to a production URL without leaving the browser. Every build runs in its own isolated cloud sandbox — a real Node environment with real installs and a real preview domain.

Front-end

Built with Next.js 15 (App Router) and React 19, styled with Tailwind CSS and Radix primitives. I implemented the token-by-token streaming chat UI, the live sandbox preview, and the file/diff explorer that stays in sync with each AI turn.

Back-end

Node.js server actions and route handlers orchestrating multi-model agents — Claude and Gemini via the Vercel AI SDK and Mastra — with MCP tool calling, per-user sandbox provisioning, a Git commit on every change, and one-click deployment.

Database

PostgreSQL (Neon) with Drizzle ORM for users, projects, message history and billing, plus Redis-backed resumable streams so a build survives refreshes and reconnects.

State management

TanStack Query for server state and AI SDK streaming state across the multi-step build, deploy and Stripe subscription flows, with Langfuse tracing for observability.

• Built coding agents on Claude Opus/Sonnet and Gemini with 14 sandboxed tools for file edits, shell and git that generate running Next.js and FastAPI/Celery apps, streaming live previews over resumable Redis-backed SSE.
• Designed the Next.js 15 platform on PostgreSQL/Drizzle with 6-tier Stripe billing, GitHub sync and 24 CI-enforced test suites.