Anirban Ghoshal
Senior Writer

Snowflake’s Cortex to bring generative AI to its Data Cloud platform

news
Nov 01, 20234 mins
Data WarehousingGenerative AI

Cortex is designed to help streamline the development of data-driven applications, use cases, AI and ML models, and foundation models from Snowpark.

shutterstock 757358335 snowflakes on snow against clouds and blue sky
Credit: ch123 / Shutterstock

Cloud-based data warehouse company Snowflake is adding more large language model capabilities and services related to generative AI, just months after releasing similar services in June.

At its annual Snowflake Snowday conference on Wednesday, the company announced Snowflake Cortex, the addition of Meta’s Llama 2 model along with other capabilities, such as new notebooks and serverless functions to aid data-driven application development.

Cortex, which is a fully-managed (serverless) service inside the Data Cloud and is currently in private preview, provides enterprises with the building blocks to use LLMs and AI without requiring any expertise in managing complex GPU-based infrastructure, the company said.

The managed service also includes serverless functions, which can be called upon using SQL or Python code, to ensure users of all skill sets can access these functions to analyze data or build AI-based applications, it added. 

The specialized functions baked inside Cortex, which are also in private preview, include a task-specific set of commands that can leverage language and AI models to boost daily analytical tasks.

“For any given input text, these models can detect sentiment, extract an answer, summarize the text, and translate it to a selected language,” the company said, adding that specialized functions include Snowflake’s existing machine learning-powered functions, such as forecasting, anomaly detection, contribution explorer and classification.

While forecasting and anomaly detection are expected to be made generally available soon, contribution explorer and classification functions are in public preview and private preview respectively.

Other general functions also packaged inside Cortex, which are also in private preview, include a text-to-SQL model and vector embedding, and search functionality based on its own foundation models and open source models.

While enterprise users can use the model to “chat” with their data, the vector embedding and search functionality allow users to contextualize a model’s responses with their data to develop an application, the company said, adding that it was adding vector as a native data type within the Data Cloud.

Cortex to take on AWS RedShift, Google BigQuery, and Teradata’s ClearScape Analytics

Cortex, according to analysts, fits Snowflake’s strategy to provide enterprises with an offering that can help streamline the development of data-driven applications, use cases, AI and ML models, and foundation models from Snowpark.

“Snowflake Cortex is a powerful way to develop generative AI applications on the data stored in the database and at the same provides an interface to do prompt engineering and retrieval augmented generation (RAG),” Sanjeev Mohan, principal analyst at SanjMo, said.

When compared to rival technology service providers, Cortex competes with companies including Google, AWS, and Teradata.

“Cortex is Snowflake’s answer to the data science capabilities that AWS Redshift draws on from Amazon SageMaker, what Google BigQuery does in combination with Google Vertex AI, or what Teradata does with ClearScape Analytics,” Doug Henschen, principal analyst at Constellation Research, said, adding that the service addresses developing AI and ML as well as generative capabilities, all from within the Data Cloud.

Other extended capabilities inside Cortex include three new features in private preview — Document AI, Universal Search, and Snowflake Copilot — all of which use either open source LLMs or Snowflake’s proprietary LLMs as their foundation.

While Snowflake Copilot helps with code generation and data analysis, Universal Search allows users to search across databases, Iceberg tables, data inside Snowflake, Native Applications, and the Snowflake Marketplace.

Document AI, on the other hand, will allow enterprises to use LLMs to extract content such as invoice amounts or contractual terms from documents and fine-tune results using a visual interface and natural language, the company said.

Cortex, according to analysts, will take at least six to twelve months to be made generally available.

Snowflake updates governance capabilities with new Horizon features

In addition to Cortex, Snowflake has added new governance capabilities to its Horizon suite, which is a built-in set of composite standards and compliance features.

These features include data quality monitoring, data lineage UI, differential privacy policies, enhanced classification of data, and other additional authorizations and certifications.  All these capabilities are currently in preview.

Snowflake is also adding a Trust Center to streamline cross-cloud security and bring compliance monitoring to a centralized place to reduce security monitoring costs. The new feature is expected to be in private preview soon, the company said.  

Horizon, according to Henschen, can be seen as a response to rivals such as Databricks and Google, who have put cataloging at the forefront with their Unity Catalog and Dataplex offerings, respectively.