The overarching message of a wave of announcements by Snowflake Inc. at its annual user conference this week will be about enabling applications to be built on its big-data cloud platform.

Among the most significant is the availability in public preview of the Snowflake Native App Framework the company introduced a year ago on the Amazon Web Services Inc. cloud. Snowflake is also adding new programmability features for developers and data scientists and taking steps to make it easier for third parties to build and sell applications in its marketplace.

The company is also joining the generative AI parade with Document AI, a large language model currently available in private preview and built on generative artificial intelligence technology Snowflake acquired with the purchase last year of Warsaw-based Applica Sp. Z o.o. Applica uses deep learning to recognize and interpret the text in documents.

Snowflake said users will be able to use the model to extract content such as invoice amounts and contractual terms easily from documents and fine-tune results using a visual interface and natural language. It plans to expand the platform’s capabilities to include more types of unstructured data over time.

The company is also updating its support for Apache Iceberg tables in a forthcoming private preview soon that will enable organizations to work with data stored in the Iceberg format either internally or in Snowflake.

Single source

In a briefing with journalists, executives emphasized that Snowflake is building upon its strategy of being a single source of data that spans multiple clouds and supports the programming environments its customers prefer to use. “We do a lot of work to make sure that Snowflake is a single product,” said Christian Kleinerman, Snowflake’s senior vice president of product. ”Once customers have integrated Snowflake into the rest of their enterprise infrastructure, a broad number of use cases light up.”

Developers can use the Native App framework to build and test applications that run natively on the Snowflake platform, taking advantage of its availability, scalability and security features and running directly within customers’ Snowflake accounts. Developers can access such tools as stored procedures, user-defined functions and table functions as well as use the Streamlit open-source Python framework to build interfaces. Telemetry features such as events and alerts for monitoring and troubleshooting are also coming.

By selling through the Snowflake Marketplace, developers can circumvent the need to set up billing systems. Snowflake said it takes a different approach to cloud marketplaces by bringing applications to the data instead of the other way around.

Customers don’t have to export or provide external access to their data to use marketplace applications. More than 25 new Snowflake Native Apps have been added to the Snowflake Marketplace and over 100 developers are building applications there, the company said.

Snowflake is also advancing its single platform to support a broader set of advanced analytics capabilities including pre-built machine learning functions for SQL users and expanding its unified governance and privacy with new data quality metrics and classification features. Both are in private preview.

New extended data programmability features have been added for data science, data engineering and application development. The company also said it’s making it easier for organizations to distribute and monetize applications at scale in its Data Cloud.

Expanded developer tools

With the release of a private preview of Snowpark Container Services, Snowflake is expanding the scope of its Snowpark developer framework to include broader infrastructure options such as access to Nvidia Corp. graphic processing units and artificial intelligence software. Snowpark Container Services users also get access to a large catalog of third-party software and applications including large language models, notebooks and MLOps tools.

Features aimed at helping organizations operationalize data and machine learning models include a public preview of a set of new Snowpark machine learning application programming interfaces for more efficient model development, a private preview of a registry that enables models to be centralized and published and a forthcoming private preview of Streamlit in Snowflake for turning models into interactive and streaming applications.

Snowpark is Snowflake’s secure deployment and processing environment for non-SQL code with runtimes and libraries specific to the Data Cloud. It enables developers to work more effectively with data in the programming languages and tools of their choice while providing their organizations with automation, governance and security guarantees.

Containerized services

Snowpark Container Services lets developers build in any programming language and deploy on a broad range of infrastructure choices in the same way they would do so in the Snowflake Data Cloud. Services can also be used as part of a Snowflake Native App to enable developers to distribute applications that run entirely in a customer’s Snowflake account.

Through a new partnership with Nvidia, Snowpark Container Services will natively support the GPU maker’s AI Enterprise platform and accelerated computing environment. Nvidia AI Enterprise includes more than 100 frameworks, pretrained models and development tools. Among the branded tools available are Hex Technologies Inc.’s notebooks for analytics and data science, machine learning platforms and features from Alteryx Inc., Dataiku Inc. and SAS Institute Inc., as well as Astronomer Inc.’s Apache Airflow-powered workflow platform.

There’s also new data content available exclusively on Snowflake Marketplace from Cybersyn Inc., a data-as-a-service company that provides proprietary economic datasets for corporations, investors and governments. Cybersyn has also launched two native applications for benchmarking e-commerce sales and monitoring the U.S. financial industry.

In a bid to highlight the company’s own continuous improvements, Snowflake is launching a Performance Index that analyzes customer workloads over time to assess performance progress. The company said the current SPI shows that query duration times have improved by 15% since last August thanks to various factors, including hardware and software improvements and customer optimizations.

“Because we charge per computer time, every time we make Snowflake faster it also implies cheaper, and we are committed to continuously improving economics for our customers,” Kleinerman said.

Photo: Robert Hof/SiliconANGLE