Back to projects
About Project
I built Code0 as an end-to-end AI development platform inspired by tools like v0 and Lovable. My goal was to empower developers and non-technical creators alike to generate, test, and iterate on full-stack Next.js web applications entirely through natural language conversations.
Behind the scenes, I orchestrated durable background execution workflows using Inngest and Agent Kit alongside isolated E2B cloud micro-VM sandboxes. The autonomous agent reasons through multi-step prompts, generates modular React components, executes terminal commands, and resolves complex runtime dependencies in real time.
To make development frictionless, I engineered an interactive split-pane workspace. Users can immediately interact with their live application inside an embedded hot-reloading iframe, explore the full project hierarchy with a recursive file tree, and examine syntax-highlighted code fragments side by side.
I backed the platform with Neon PostgreSQL and Prisma ORM to preserve prompt histories, project states, and code revisions across sessions. Combined with Clerk authentication and a custom warm aesthetic, Code0 delivers a production-grade, reliable, and approachable AI software creation experience.
Challenges
- [-] Standard serverless timeouts disrupted multi-step agent reasoning. I resolved this by utilizing Inngest's durable workflows to manage agent execution steps, retries, and state transitions reliably.
- [-] Provisioning fast, secure environments for untrusted code execution was tricky. I leveraged E2B cloud micro-VMs to run commands, stream logs, and preview Next.js apps safely.
- [-] Keeping client file trees synchronized with dynamic sandbox file writes was challenging. I designed a PostgreSQL schema storing versioned code fragments for instant UI hydration.
- [-] Embedding untrusted sandbox web apps triggered cross-origin and security barriers. I implemented secure HTTPS iframes paired with automated port forwarding to deliver seamless live application previews.