Stackex

Stackex is an AI-powered platform that builds Expo / React Native mobile apps from natural-language prompts, with live per-user dev environments in the browser. Founding engineer and tech lead — built it from zero, with a second engineer joining the team later.

The platform has two halves working together: the AI agent layer that writes the code, and the sandbox infrastructure that runs each user's generated project live in the browser.

• Built per-user sandbox infrastructure on GCE — every user gets an isolated Docker container running their React Native project including the Metro bundler and dev server.
• Designed the reverse-proxy routing layer using Nginx that maps each browser session to the correct sandbox through a single domain, plus a Redis-based route cache so request resolution stays fast at scale.
• Built the AI agent execution layer with two modes: direct execution for small edits (up to fifteen iterations), and a thinking/planning/execution pipeline for larger features — where each plan step gets its own iteration loop with independent LLM context.
• Implemented the agent's tool system — file read, file modification, and in-sandbox command execution — with full tracing via Langfuse for end-to-end debugging of the agent's reasoning chain.
• Wired a browser-based terminal into sandboxes with a restricted command surface to prevent escape or cross-tenant impact.
• Built multi-tenant lifecycle management — provisioning containers, mapping ports, short-lived token authentication, and firewall rules at the container level to prevent cross-tenant access.

Platform Architecture

Per-user sandbox infrastructure on GCE with isolated Docker environments and AI agent execution layer.