About Snorkel AI
Snorkel AI was founded in 2019 by Alex Ratner, Braden Hancock, Henry Ehrenberg, and Paroma Varma, spinning out of the Stanford AI Lab, where the founding team had developed Snorkel, an open-source research project addressing a persistent bottleneck in building machine learning models: the need for large volumes of accurately labeled training data, meaning data where the correct answer or classification has already been determined, typically by having human annotators manually review and label each individual data point, a slow, expensive, and often significant bottleneck in developing machine learning models for specialized enterprise use cases where large, pre-existing labeled datasets don't already exist.
Snorkel's core technical approach, called "programmatic labeling" or "weak supervision," lets data scientists and domain experts write labeling functions, essentially rules or heuristics encoding their domain expertise about how to identify correct labels, which Snorkel's underlying algorithms then combine and reconcile across potentially many different, sometimes conflicting labeling functions to generate a probabilistically weighted training dataset, an approach that can dramatically reduce the amount of purely manual, one-by-one human data labeling required to build a training dataset, particularly valuable for specialized enterprise use cases involving domain-specific data, such as legal documents, medical records, or specialized financial data, where hiring and training human annotators with adequate subject matter expertise to label data accurately can be particularly expensive and slow. The company commercialized this research into an enterprise platform that lets organizations apply programmatic labeling techniques to their own specific, often data-scarce or highly specialized machine learning use cases, and has increasingly positioned its technology as relevant to fine-tuning and evaluating large language models for enterprise-specific use cases, since even powerful general-purpose language models often benefit significantly from being fine-tuned or evaluated using labeled data reflecting an enterprise's specific domain and use case requirements. Snorkel AI's customers span large enterprises across regulated industries including financial services, healthcare, and government seeking to build specialized AI models more efficiently. The company has raised substantial venture funding from investors including Addition, GV, and Lightspeed Venture Partners, reaching a valuation exceeding $1 billion.
