Snowflake Trifecta: Data Extraction, Time-Series Forecasting, and Cortex LLMs

At the Mallet Brothers Show

I first discovered Snowflake in 2018, after wrestling with Vertica's instability and Redshift's sky-high scaling costs. As organizations migrated to the cloud, Snowflake was an obvious choice as a modern data warehouse. It cleanly decoupled compute from storage, integrated seamlessly into AWS S3, and added features like Time-Travel and Zero-Copy Cloning. Those steeped in Oracle and SQL Server data warehouses were astonished at how effortlessly Snowflake scaled.

Ironically, by becoming the industry’s de facto ‘Data Warehouse Platform,’ Snowflake ceded the machine-learning spotlight to Spark-based rivals. Databricks rode its Spark-first, notebook-centric approach to dominate the ML conversation. At the same time, data catalog and core governance remained afterthoughts, even as Snowflake quietly expanded into a complete data cloud platform that embraced Python and machine learning.

I'll confess: I didn't keep up with how fast Snowflake is taking their already solid platform and building end-to-end machine learning and AI capabilities directly into the warehouse. The fundamental idea makes sense: Once you've decoupled compute from data and have a robust compute platform, it's an easy extension to build more advanced logic into that pipeline.

I just had the pleasure of porting the Inventory Analysis Case Study by PwC to use Snowflake within the notebooks environment. I was amazed at how easy it is to integrate ML and AI in a few steps.

1. Extract inventory data directly from a SALES table into Pandas for analysis

2. Forecast future stock levels with the built-in ML.FORECAST time series function.

3. Interpret results and build subsequent actions using plain language from an LLM using Cortex.

   
   

Together, these capabilities are a trifecta for building the types of AI applications that will dominate enterprise development in the next few years.

Here is the code for your perusal. One of the bonus features I explored was Notebooks integration into Git, so I'm managing my branches and GitOps directly from Snowsight.

• https://github.com/timowlmtn/azrius-snowflake