It all started when our founder, Zheng, started dabbling with machine learning early in his career and eventually landed at Cruise AI's Perception Platform & Machine Learning Accelerators (MLA) team in 2019. It was there where he and his peers learned how challenging it was to even deploy models that our researchers have developed.
Everything from managing datasets, to validating our models to not just be functionally correct, but fit within the system performance limits of our hardware constrained autonomous vehicles. The model deployment cycles were always longer than we'd like them to be.
Many of these deployment tools / solutions weren't readily available, so we had to build them from scratch. Imagine a world where Stripe doesn't exist and having to build billing for a modern software stack. That's how painful it was and still is.
Zheng has heard from several ML engineers at other autonomous vehicle companies who also face similar pain points. Some came from different industries with a business need for more optimized models but can't justify the R&D costs for better deployment tooling. Wouldn't it be nice if everybody could benefit from a solution focused solely on this vertical? What if Cellulose could fill this gap?