For H2O.ai, a machine learning company, democratizing generative AI is not an empty motto, but a mission — one that requires action. And action depends on getting models, automated tools and analytics into the hands of people who can use them to experiment, iterate and create new uses for AI technology. 

H2O.ai’s primary goal is to simplify customer access to data for AI model training and inferencing, while safeguarding its customers’ data privacy and reducing data movement. To that end, H2O.ai offers its state-of-the-art ML engine and a slew of additional gen AI capabilities for making and fine-tuning custom GPTs — including the H2O Driverless AI AutoML engine; H2O LLM Studio; and starter packs for machine learning, gen AI LLMs and predictive modeling — as Snowflake Native Apps on Snowflake Marketplace. 

The Snowflake Native App Framework integrated with Snowpark Container Services (currently in public preview) allows H2O.ai to simplify and accelerate the process of getting AI and ML models up and running in customers’ environments.

“Snowflake Native Apps integrated with Snowpark Container Services smooths out that entire process,” says Luis Boldizsar, Channel Manager, North America at H2O.ai. “It lowers the bar to entry and makes it easy for customers to consume an app inside of Snowflake, to access the technology and use it to build new data insights.”

H2O.ai saw the potential of Snowpark Container Services during early engineering briefings with Snowflake. A fully managed container offering, Snowpark Container Services presented a way for H2O.ai to deploy its solutions directly inside Snowflake, close to the data it interacts with. Snowpark Container Services encapsulates the complexity of the orchestrations, integrations, dependencies and associated back-and-forth management, making it simpler to deploy and removing the need to move data between environments outside of Snowflake.

Snowflake Native Apps provided the next piece, smoothing out infrastructure problems and making it “incredibly easy for a user to start consuming an application inside of Snowflake,” says Boldizsar. The H2O.ai team can now have one of their native apps and a product up and running inside a customer’s environment in 20 to 30 minutes instead of several days.

“That type of timeframe is a major benefit to customers,” says Boldizsar, noting that H2O.ai has been able to minimize the “dead time” when teams are waiting for things like infrastructure availability or moving into production. Because Snowflake Native Apps run within the customer’s account, they may benefit from Snowflake governance capabilities and customers can vet, approve and onboard new apps more quickly. 

Snowflake Native Apps integrated with Snowpark Container Services also allows H2O.ai to include artifacts (which are auto-generated) along with the code. This means that when a data scientist passes a model to a data engineer, they also get additional useful information, such as an example of how to invoke it in SQL or a piece of code and example of how to consume it in Python.

“It might seem like a small thing, but I know customers that have spent an hour or more during a handoff to explain things like how to call a model or what data to pass,” says Boldizsar. “When we reduce the time for all those tasks, it helps customers get models into production and start to see value from using these models and this data. And then it starts to snowball — because they can do a task faster and in a secure way, they can now fit in another project.”

H2O.ai created an innovative process by using templates to speed development of its apps. By incorporating Snowpark Container Services into its Snowflake Native App templates, H2O.ai can eliminate complex container-loading processes and take advantage of the Snowflake Native App Framework to simplify the installation process.

Additional benefits of H2O.ai’s template emerge when it’s time to distribute updates and new releases. Decoupling the image development from the Snowflake Native App code significantly reduces the overall development time and allows for swift updates, so both H2O.ai and its customers can keep pace with frequent releases of new capabilities in the applications. It’s a significant advancement in ML and data processing that allows H2O.ai to provide exceptionally efficient, scalable and user-friendly solutions. 

The fact that Snowflake Native Apps run in the customer’s environment and do not require moving data or providing external access to it is especially beneficial for H2O.ai’s customers in the highly regulated financial services industry. At the same time, customers cannot rummage through the app’s code and make variations that — while well-intentioned — may produce an unexpected side effect. 

“With Snowflake Native Apps, we are able to install our template in the way we set it up and the way we know it works. It gives us confidence that our apps will work as we intended,” says Boldizsar. “That same capability protects the customer, too, because they know we can’t see their code or their data either. 

“Sometimes a fully managed service provides more capability, but it’s also more complex for us and our customers to run and collaborate on. A simplified platform, like what Snowflake provides, removes some of that back and forth” and can be easier to manage, says Boldizsar.

The Snowflake Native App Framework's target release directive feature allows H2O.ai to provide different versions of its apps so customers can test them directly. H2O.ai puts the app version number into the title, so a customer can choose to keep and patch their current app to remain compliant with regulations that require specific engine versions, or upgrade and install the newest version.

Those interested in H2O.ai’s gen AI apps built with the Snowflake Native App Framework integrated with Snowpark Container Services can try them for free on Snowflake Marketplace. 

Meanwhile, the H2O.ai team continues to explore the potential of Snowflake Native Apps with Snowpark Container Services, experimenting with packaging and looking into different configurations that could give users the full range of ML and AI capabilities in one service. 

“That’s a good example of democratizing access to AI: enabling data scientists to build new models and tackle problems they previously haven’t tried to address, just by making it super simple to access the tools to make it happen,” says Boldizsar.