SNORKEL AI

Alex Ratner, Chris Ré (a Stanford CS professor), Henry Ehrenberg, and Braden Hancock spent years at Stanford researching how to build AI systems with less labeled data. Traditional ML required massive hand-labeled datasets, which were expensive and time-consuming to create.

They developed "data programming" — an approach where domain experts write labeling functions (simple rules and heuristics) that automatically generate training labels. The open-source Snorkel project became hugely popular in the ML community, and they founded Snorkel AI in 2019 to commercialize it.

Enterprise SaaS. Snorkel AI charges annual license fees for its data-centric AI platform.

Pricing is based on users, data volume, and features. Revenue comes from enterprise contracts with large organizations — government agencies, financial services, and tech companies — that need to build AI systems quickly without massive manual labeling efforts.

Snorkel Flow (data-centric AI development platform). Programmatic Labeling (write rules instead of manually labeling).

Data Slicing and Exploration. Model Training and Evaluation.

LLM Fine-Tuning (preparing data for foundation model customization). Integration with major ML frameworks.

Stanford credibility and government contracts. The Stanford pedigree gave Snorkel AI immediate credibility in both academic and enterprise AI communities.

Government and defense contracts (DARPA, DoD) provided early revenue and validation. The company then expanded to commercial enterprise clients.

The rise of large language models (GPT-4, etc.) that require less task-specific labeled data. If foundation models can perform tasks with few-shot learning, the need for large labeled datasets — Snorkel's core value proposition — may diminish over time.

The company has pivoted to focus on LLM fine-tuning and data preparation for foundation models.

Backed by Lightspeed Venture Partners, Greylock Partners, GV (Google Ventures), and BlackRock. Lightspeed led the Series C in 2022.