Snorkel Custom

Accelerate data-centric AI development

Snorkel Custom is a hands-on accelerator for enterprises using large-language models (LLMs) to build AI applications. It pairs the Snorkel Flow AI data development platform with a team of Snorkel AI experts to deliver and serve fine-tuned and cost-optimized models—along with a co-developed benchmark for evaluating model performance against specific goals and objectives.

Get started

Read the announcement

Industry leaders rely on Snorkel for accelerated AI development

Benefits

The power of data-centric AI

The key to getting LLMs to deliver production-quality results lives in enterprise data. Snorkel built the world’s first and only programmatic AI data development platform for labeling, curation, and development of enterprise data to make LLMs deliver higher accuracy and optimized performance. Snorkel Flow is used by enterprise customers, including 7/10 top US banks and Fortune 100 companies across 10+ verticals, to label and curate data for training and tuning models.

Snorkel Custom accelerates production AI with the unique technical differentiation from Snorkel Flow and the extensive experience of our research team.

Snorkel experts collaborate directly with your internal data scientists and SMEs to support the entire path to production, ensuring best practices for data development, model tuning and evaluation, and a supported implementation of the Snorkel Flow platform as the core infrastructure for data-centric AI.

Higher accuracy with specialized models

Snorkel Custom engagements produce specialized LLMs tuned and aligned on an enterprise's data, for production accuracy on specific use cases.

Accelerated delivery of current & future models

When inputs and business objectives change, LLMs need to be tuned on new data. Snorkel's programmatic data development platform makes it easier to update data inputs and tune models.

Reduced data labeling & model serving costs

Snorkel Flow's programmatic approach to data development reduces the time and cost of data labeling by 10-100x, and creates more accurate models that are substantially smaller and far more cost-efficient.

Zero model vendor lock-in

Snorkel Custom engagements are built using Snorkel Flow to tune and adapt any base LLM model, open or closed source. The tuned models can be continuously hosted and maintained by Snorkel, or transferred to a fully self-serve model.

Process

Snorkel Custom
engagement and delivery

A proven, data-centric process for production AI

Custom LLM benchmark and evaluation

Collaborate to build a custom LLM benchmark off ground truth data and use-case specific metrics, used to achieve production quality performance.

Create training and validation data

Trained experts will import custom data into Snorkel Flow, and collaborate with your internal SMEs to label and curate training data.

Fine-tune and align models

The Snorkel team will tune and adapt foundational LLMs to optimize performance and accuracy against the custom benchmark.

Cost-optimization and distillation

Snorkel experts distill a base model into smaller models, which perform specialized tasks with high accuracy and are much more economical to serve.

Benchmark and production deployment

Collaboratively run the custom LLM benchmark on all new models to ensure they meet accuracy requirements, and deploy the optimized model to a production environment

Real-world success

Wayfair launched an initiative to extract information from millions of product images, but faced challenges due to poor training data, years of effort estimated to manually label data and a foundation model that made inaccurate, high-confidence predictions.

Snorkel collaborated with Wayfair to develop a workflow combining data preprocessing, labeling functions and foundation models—resulting in quality training data and high model accuracy.

20+

point improvement

300x

faster development

46

production models

See how it’s done

Upcoming webinar

How Wayfair is transforming customer experiences with data-centric AI

In this upcoming webinar, Snorkel AI experts will demonstrate how to programmatically label data, use it to build a training dataset and fine-tune an open source model – all with Snorkel Flow.

Register

Vinny DeGenova

Associate Director of Machine Learning @Wayfair

Daniel Xu

Product Lead @ Snorkel AI

Kieran Kavanagh

Principal Architect, Retail @ Google Cloud

AI leadership

Build internal AI competency

Snorkel was founded by AI researchers who pioneered data-centric AI, and who continue to advance state-of-the-art techniques for adapting LLMs to specialized tasks. Snorkel researchers have published over 100 peer-reviewed research papers, with special recognition at events such as NeurIPS, ICML, and ICLR.

Be at the forefront of AI

The long-term objective of Snorkel Custom is to help enterprises grow the internal AI competency required to sustain continued AI development, apply it to emerging use case and create competitive advantages.

Dive deeper with these resources

Snorkel Custom