In the enterprise, it’s common for data to reside on three separate platforms: transactional databases, data warehouses and machine learning (ML) environments.
Most artificial intelligence (AI) applications are built on historical data that has been loaded into ML environments for data mining. So you’re looking in the rearview mirror to figure out what has already happened, he said. If you want to make decisions in the moment, you need an operational data store, like a traditional database management system or, for more scalability, a NoSQL data store. Lastly, ML brings its own algorithms, data science workbenches and modeling workflows.
Currently, it’s up to developers to duct tape all of these workloads together.
Splice Machine solves all of those problems and uses SQL as its basis. Its AI data platform seamlessly puts all of these components together, replacing old data platforms with one that is scalable up to petabytes, allows for instant analysis through the ML platform and makes the data instantly available to end users.
This allows businesses to modernize purpose-built applications that they’ve spent thousands of dollars on. These customizations are often key differentiators in their marketplaces, Zweben said, and they are understandably loathe to give up this customization to modernize. But it’s like they have their feet in cement because those highly customized applications can’t be brought into the new tech stack.
The Splice Machine operational AI platform allows customers to keep all of their customization, eliminate cumbersome and time-consuming extract, transform and load (ETL) processes, add ML components to their data analysis and gain the advantages of modernizing their stacks all in one fell swoop.
For example, Splice Machine has been working with an insurance company that had built ML algorithms for fraud detection. But by the time the claims data had been extracted from the business system, run through the ETL process, and loaded in the ML models, it had changed so many times that the data was stale at the ML location, and so the fraud detection scores were outdated and therefore inaccurate.
By adding Splice Machine to the stack, this company was able to operationalize those existing Spark-based business models and inject them directly onto the business system, allowing real-time fraud detection.
“That is the epitome of having an operational AI data platform,” Zweben said. It can run the business system with the operational capabilities but also have the analytical capabilities and the ML models resident on that same platform.
Listen in to hear more about business use cases for this operational AI platform, the steps to the modernization journey, Zweben’s 20-year career journey through AI starting at NASA and how that all informed his building of Splice Machine.
In this Edition:
3:54: Splice Machine on the stack.
8:44: Integrating Splice Machine.
17:51: Taking data to the next level.
22:17: Splice Machine history.
29:40: Splice Machine’s appeal to CFOs and financial departments of data and data platforms.
32:47: Recommendation for learning material.