We build autonomous
systems that work.

The lab began where most engineering does: in a room full of noisy machinery. While working on AI diagnostics for building systems — chillers, air handlers, the hidden infrastructure that keeps cities breathing — a pattern emerged. The data was rich. The tools were not.

So the tools were built.

PyEmits came first. Sugar candy for data scientists — a Python library that made time-series manipulation feel effortless. No ceremony, no boilerplate. Just clear APIs and fast insights. It was the first expression of a principle that would guide everything that followed: complexity is the problem. Tools are the answer.

Then came PyConn — a uniform connection layer across databases, warehouses, and object stores. The insight was simple: engineers shouldn't have to learn a new dialect every time they switch storage backends. One interface, one mental model, no friction.

Years building data platforms at scale — ETL pipelines humming across Hong Kong's startup ecosystem, MLOps infrastructure running on DGX SuperPODs, growth strategies powered by clean analytics — sharpened a deeper conviction. The best infrastructure disappears. The best tools feel like extensions of thought.

promptv was the turning point. As LLMs matured from toys into tools, the same old problem resurfaced: versioning, collaboration, provenance. So promptv gave prompt engineering the one thing it was missing — git-like version control. A developer-first tool for a developer-shaped problem.

Today that same instinct drives everything at the lab. Multi-agent coordination. Persistent memory layers. The scaffolding that turns a prompt into a production system. The question is always the same — what tool would make this effortless? — and the answer is always code.