Snorkel AI

Turning domain knowledge into training data

Snorkel AI’s 2020 launch described using programmatic labeling and weak supervision to build training datasets through Snorkel Flow. The founders said the company spun out of Stanford AI Lab in 2019. The commercial company is distinct from the earlier academic project and papers. [1]

A shipped platform and a broader data business

The July 2020 announcement brought Snorkel Flow out of stealth after a year of building and deploying it. The relevant milestone is the platform, rather than funding alone or an automatic Stanford ownership relationship. [1]

Its current services include expert-authored data, preference rankings, evaluation rubrics, graders and runnable environments. These describe what the company supplies to AI development. Whether a dataset improves a particular model still depends on a measured experiment, rather than the vendor’s general claims. [2]