Snowflake’s Data Cloud Summit 2024 was full of exciting announcements across all workloads, with a strong emphasis on AI/ML innovations. From Generative AI and Large Language Models (LLMs) to Predictive AI/ML and MLOps capabilities, it’s evident that the Snowflake AI Data Cloud continues to prioritize advanced AI applications.
In this blog, we will focus on AI capabilities and interfaces – we have another post dedicated entirely to developer experience and UI features for data scientists!
Here’s a quick rundown of our favorite AI announcements from the event!
Cortex Serverless Fine-tuning
Snowflake Cortex is already generally available (can you believe the preview was announced just last winter?), including models from Arctic, Mistral, Reka, and Llama3. Snowflake has made another leap forward by introducing the Cortex Serverless Fine-Tuning API.
This allows users to fine-tune LLMs to their enterprise context using a simple low/no-code interface by pointing it at an existing dataset. Models fine-tuned via this interface can be managed with the Snowflake Model Registry, enabling a unified RBAC security model for both LLMs and the data used to train them.
These enhancements illustrate Snowflake’s ability to evolve Cortex from model serving to model training, reinforcing its position as a leading platform for enterprise Data & AI initiatives.
LandingLens Native Application from LandingAI
With its robust capabilities for training, tuning, and deploying domain-specific Large Vision Models, LandingAI leads the industry in data-centric Visual AI for enterprise use cases.
At Summit, Snowflake and LandingAI furthered their partnership by announcing that LandingLens is now available as a Native Application on Snowflake. Customers can now deploy a secure, single-tenant LandingLens instance within their Snowflake account with just a few clicks.
This seamless integration offers an easy pathway to adopting cutting-edge Visual AI technology, eliminating the security and governance challenges of sharing images and intellectual property over the public internet.
For more details, check out our LandingAI & phData partnership announcement!
Cortex Copilot API
The announcement of the Cortex Copilot API significantly enhances Snowflake’s capabilities for building complex data-driven AI applications. Cortex Copilot was originally launched as an end-user tool to assist in writing SQL and enabling natural language queries, but Snowflake has now exposed that functionality as programmatic API.
This allows developers to leverage Copilot’s SQL-writing capabilities, which are enhanced by its deep integration with semantic metadata stored within Snowflake. For developers who need even more flexibility, Snowflake has gone a step further in allowing API clients to specify custom semantic metadata, such as dbt metadata.
This evolution opens up new possibilities for integrating sophisticated, metadata-driven query generation into data workflows, as detailed in our original Talk to Your Data solution.
Cortex Search API
Cortex Search stands out for its powerful hybrid search technique, combining classic semantic search with AI-driven vector embeddings. Initially created as as a universal catalog search feature for data (and metadata) in Snowflake, its core functionality has now been extended for programmatic use Cortex Search API.
This API can power Retrieval-Augmented Generation (RAG) applications and even enhance those applications relative to architectures that use vector embeddings alone.
By leveraging both semantic and vector search capabilities, Cortex Search provides a robust solution for complex data-driven applications for organizations with a large data estate. The Cortex Search API adds yet another building block that simplifies the development of custom enterprise applications, similar to the ease of use offered by the Cortex Copilot API described above.
Snowflake Feature Store
Feature stores are an important technology category that we’ve long been following. The Snowflake Feature Store is now generally available, and it’s immediately a compelling player in that category. Like any feature store, it exposes features for consumption but stands out with its simple interface for defining and materializing transformations.
Additionally, the lineage of these transformations is displayed through a sleek new UI. At phData, we’re leveraging the Snowflake Feature Store for both Generative AI (such as RAG on structured data) and for Predictive AI/ML and MLOps platforms.
Takeaways
The prolific volume of advancements unveiled by Snowflake at Summit in the realm of AI/ML this year is truly striking. Witnessing Snowflake’s swift evolution beyond traditional data warehousing workloads and expanding its reach to embrace ML applications across the vast data volumes within its secure platform is very exciting for us at phData.
While some of these features may still be in Public or Private Preview, as Snowflake’s 2024 partner of the year, phData has had the privilege of collaborating with these innovations prior to their official release.
Are you excited to dive deeper into these latest developments or seek guidance on optimizing success with Snowflake? Rest assured, we stand ready to address any inquiries you may have!
