H2O AI and Snowflake
H2O and Snowflake bring AI to the data instead of moving data out. H2O eScorer deploys models into Snowflake as user-defined functions for real-time and batch scoring, cutting scoring elapsed time 30x (2024). H2O apps in Snowflake Marketplace run on Snowpark Container Services, so teams train and score predictive models and use generative AI inside their own Snowflake accounts.
Partnership overview
Generative AI as Snowflake Native Apps
In June 2024, H2O launched generative AI capabilities as Snowflake Native Apps on Snowflake Marketplace. Three starter packs run inside your Snowflake account on Snowpark Container Services (SPCS).
H2O Predictive Modeling Starter Pack
H2O Driverless AI and H2O eScorer in an SPCS bundle to train and deploy models.
H2O GenAI LLM Starter Pack
h2oGPT with H2O LLM Studio to tune and use LLMs in SPCS.
H2O Machine Learning Starter Pack
H2O-3 and H2O eScorer to train and deploy classical machine learning models in SPCS.
Distributed training across Snowflake clusters
H2O is the only AI/ML solution that can run across multiple clusters on Snowflake. Using Snowpark Container Services, H2O lets Snowflake users run distributed training on massive datasets.
Key benefits
- Train more frequently
- Use the most up-to-date data
- Achieve more accurate results
- Lower TCO
Scoring in Snowflake removes data movement and extra processing steps. That cuts cost and management effort, and brings a 30x reduction in scoring elapsed time. Snowflake users manage machine learning in the interface they already know, with a simpler workflow and real-time scoring that improves model performance.
“Integrating Driverless AI and eScorer within Snowflake's cutting-edge container services and native apps has revolutionized our approach to data analytics. Collaborating closely with H2O.ai, our valued partner rather than just a vendor, we've witnessed the power of bringing intelligence to the data, rather than the traditional approach of moving data to intelligence. This shift has not only enhanced efficiency and performance but has also opened doors to a realm of advanced analytics possibilities. Together, we're not merely implementing solutions but co-creating transformative strategies that drive our business forward. This deep-seated partnership ensures that we harness the full potential of H2O's technologies to deliver unparalleled value to our customers and stay at the forefront of the ever-evolving landscape of data analytics.”
Jeffrey Vagg
Chief Data and Analytics Officer, North American Bancard (NAB)
Key features of the native apps
- Integration. Use generative AI models directly within Snowflake Native Apps, with no complex integrations or specialist expertise needed.
- Data enrichment. Generative AI apps enrich data by generating new columns from advanced calculations, adding insight for better decisions.
- Predictive AI. Combined with traditional predictive models, generative AI answers sophisticated questions and generates content from existing data stores.
“At Yieldmo, we rely on cutting-edge data processing to power our advanced machine learning and AI applications in the advertising technology industry. Predictive modeling is at the heart of the value we deliver to our clients, enhancing everything from predictive audiences to high-performing inventory and ad creative. By leveraging significant big data partnerships with platforms like Snowflake and H2O.ai, we deliver enhanced value and proven results. We have employed H2O's Driverless AI for years to train sophisticated machine learning models. With the recent introduction of H2O eScorer on Snowflake’s container services, we can scale the predictions made by these models to a new level of volume and accuracy. In early testing, we reduced prediction time by more than tenfold while cutting costs by half compared to our existing system. These spectacular results were achieved by eliminating the necessity of external data transfer and effectively utilizing more computational resources.”
Eric Shiffman
VP of Product Marketing, Yieldmo