Snowflake launches LLM-driven Document AI and more at annual conference 

Snowflake is making another generative AI push. Today at its annual conference, the data cloud company announced Document AI, a new large language model (LLM)-based interface that allows enterprises to quickly extract value from their barrage of documents. The move marks a major development for the data giant — which started off with a focus on structured data — and provides an easy way to mobilize useful unstructured information that often remains scattered across silos.

“We’re unlocking a new data era for customers, leveraging AI and eliminating silos previously bound by format, location and more to revolutionize how organizations put their data to work and drive insights with the Data Cloud,” said Snowflake SVP of product Christian Kleinerman.

The Montana-based company also debuted new iceberg tables, ML-powered SQL functions and cost optimization tools at the event.

New way to mobilize unstructured information

Back in September 2022, Snowflake completed the acquisition of Poland-based Applica, an AI platform for document understanding. This technology is now powering the new Document AI.

As the company explains, all an enterprise user has to do is detail what they want in natural language. The interface automatically processes that query to extract the required content and analytical insights from the document in question, be it an invoice, contract or something else.

“Customers will see an end-to-end experience where they will able to have documents in Snowflake and ask structured questions from those documents — like what’s the name of the employee, what’s their address or the total value in the invoice,” Kleinerman explained in a press briefing. “This will trigger the system to take the documents, which are unstructured files, and convert them into structured data.”

Once available, this converted data could be used for traditional analytics, BI or other downstream ML processes, the SVP added.

At the core of Document AI is Applica’s purpose-built, multimodal LLM that processes language queries to provide outputs. Snowflake said it is working to expand this system to cover more types of unstructured data, but has not said what would be next. To note, unstructured data is very comprehensive and can include images, text files, videos and much more.

The move could play a major role in Snowflake’s growth story, as IDC estimates that more than 90% of the world’s data will be unstructured over the next five years.

Other happenings at Snowflake Summit 2023

Beyond Document AI, Snowflake announced updates for Iceberg tables and ML-powered SQL functions for its data cloud.

The former, Kleinerman noted, will help enterprises converge native and external tables for Iceberg into a unified table type, making it easier to extend the value of data cloud to Iceberg data, while the latter brings machine learning (ML) to the hands of non-technical data users, allowing them to easily handle use cases like forecasting or anomaly detection.

Finally, the company debuted Snowflake Performance Index, an aggregate metric to quantify query performance for enterprises as well as two new cost optimization tools: Budgets and Warehouse Utilization. 

Budgets will set the threshold for maximum spending on compute during a specific period of time and issue alerts when the limit is about to be breached. Meanwhile, Warehouse Utilization will give enterprises visibility into how well-utilized their compute clusters are, allowing them to downscale when possible and save money.

Snowflake Summit runs from June 26 to 29 in Las Vegas.