AI Lab
An AI-powered online IDE for conducting coding assessments, giving candidates a VS Code-like development environment with an AI agent, terminal, sandboxed execution, and live preview while allowing evaluators to assess their work.

My contribution
Solely led the architecture, design, and development of AI Lab from scratch. Designed the browser-based development environment, local virtual filesystem, AI agent integration, terminal and sandbox architecture, model abstraction layer, and live preview system.
Skills & technologies
Things worth knowing
- Since launch, AI Lab has handled 500+ real-world assessment sessions every week.
- Designed and built an online AI coding assessment environment from scratch with a VS Code-like interface.
- Built an AI agent experience where candidates solve programming problems by interacting with the agent through a dedicated chat interface.
- Created a browser-local virtual filesystem using IndexedDB so candidate code could be generated, edited, and persisted locally.
- Developed and published the `vfs-idb` NPM package to provide a reusable virtual filesystem layer backed by IndexedDB.
- Designed the filesystem so the editor, AI agent, terminal, and execution environment could work with the same project files.
- Integrated Gemini, OpenAI, and Claude through separate model connectors behind a common architecture.
- Added terminal, compiler, and sandbox capabilities for programming assessments requiring actual code execution.
- Integrated E2B for isolated sandbox execution.
- Built live preview support for UI and web-based coding assessments.
- Designed the platform around assessment constraints including time limits, AI chat limits, and evaluation requirements.

