Skip to content
Harjot Singh Rana

Full-Stack & AI Product Engineer

All projects

Moonshift

AI app builder that turns prompts into deployed full-stack applications. Multi-agent generation pipeline, SSE real-time observability, and production usage by 1,200+ users generating 800+ apps.

Live
Screenshot of MoonshiftScreenshot of Moonshift (dark mode)

Case study

Situation

Moonshift helps founders and developers turn a natural-language idea into a launched product on their own infrastructure. A successful run had to produce more than a demo: a real app, a GitHub repository, a Vercel deployment, and a launch kit.

Constraint

Multiple agents and external side effects had to work together without hiding cost, reliability, or publishing decisions from the user. Deployment and social publishing needed explicit approval gates.

Stack

TypeScriptPythonNext.jsPostgreSQLNode.jsOpenAIDocker

What I built

  • Rebuilt the multi-agent pipeline around research, planning, contracts, validation, and bounded fixer loops so generated code had a checkable path to production.
  • Added live SSE run observability for phase progress, agent state, generated assets, deployment, and the human approval gate.
  • Hardened the ship path with cost caps, resumable runs, and GitHub, Vercel, Turso, Playwright, and Docker integrations.

Outcome

Raised end-to-end generation success from 46.2% to 76.7%, cut median prompt-to-live time from 22.1 to 14.9 minutes, and supported 1,200+ users generating 800+ apps. The run dashboard streamed 7.3M+ SSE events across 2,000+ runs.

Role
Founding Engineer
Timeline
April–July 2026

Technical detailBun + TypeScript monorepo; Next.js dashboard; orchestrator-v2; SSE; Drizzle/Turso; Vercel; GitHub; Playwright; Docker.

Discuss a similar product

Key features

  • Prompt-to-deploy full-stack app generation
  • Multi-agent LLM orchestration pipeline
  • Real-time SSE observability at multi-million event scale
  • Production use: 1,200+ users, 800+ generated apps