Dremio enables you to manage changes more effectively than any other data warehouse platform. There are two things that come into play. One is data lineage. If you are looking at data in Dremio, you may want to know the source and what happened to it along the way or how it may have been transformed in the data pipeline to get to the point where you're consuming it. There's another thing called data providence. They're tied together. Data providence allows you to go back and recreate the data at any particular point in time. It's extremely important for compliance and governance issues because data changes all time. How did it change? What was it three days or months ago? You may have made some decisions based on data that was three months old, so you might need to revisit those. It's essential for things like machine learning and deep learning, where you are generating AI models off data. When the model stops working or doesn't work as expected, you need to figure out why. You have to go back and adjust the datasets used to train the model. We do that through an open-source project called Nessie, which is their basis for providing data lineage and data province capabilities. It's super powerful. Arrow is another open-source project for storing data in memory and performing data query operations. Data sits on a disk in one format. If you want to do anything with data, you have to load it into your computer and put it into memory so you can work with it. Arrow provides a format in memory that enables the whole library to perform various operations on that data. Every vendor has its own way of representing data in memory. They've latched onto an industry standard and developed it so it's open. Now people can use the exact same format in memory to do operations and use the library set to perform functions on data. New developers can decide if they want to develop their own memory format or use one that's already there. Data transfer is a massive problem when you're working with large datasets, doing advanced analytics, and trying to train machine learning or deep learning models. What happens often is companies downsample their data sets to do training on models because transferring and managing data on a deep learning or machine learning platform is too much.
