AiThority Interview with Dr. William Bain, CEO and Founder of ScaleOut Software

Dr. William Bain, CEO and Founder of ScaleOut Software, shares more about their latest product releases, trends defining digital twin applications for real-time monitoring, and more about the future of AI in this Q&A: 

Hi Dr. William, tell us about yourself and the story behind ScaleOut Software.

I have a career focus in the field of parallel computing that dates to my graduate work at Rice University in the late 1970s. I worked at Bell Laboratories and Intel’s supercomputing division prior to starting a software company in 1997 that was acquired by Microsoft. After integrating our company’s technology into Microsoft Windows Server, I decided to start ScaleOut Software in 2003 with the goal of using supercomputing techniques to tackle the challenge of managing (storing and analyzing) large volumes of fast-changing, live data. Our company has delivered a series of products to the market that are built on a core, parallel computing architecture for fast, scalable data management of live data combined with maximum ease of use. Our latest product, ScaleOut Digital Twins, lets our customers build streaming analytics applications at scale using a breakthrough approach to introspecting on live data. It represents the culmination of two decades of technology development, and we are very excited to explore the many industries’ use cases that can benefit from it.

Share with us the top highlights of your latest enhancements and product releases.

With Version 4 of ScaleOut Digital Twins, we are delivering our first integration of generative AI and real-time monitoring using digital twins. In a previous release, we integrated machine learning (ML) with digital twins so that streaming analytics applications can more easily detect subtle anomalies in telemetry using trained ML algorithms. Generative AI takes these capabilities to the next level by assisting managers in aggregating and visualizing information generated by digital twins, including analysis performed using ML. It also can monitor analytics results aggregated by our platform for anomalies so that managers do not have to continuously watch dashboards for emerging issues.

In this release, we have also added automatic retraining for ML algorithms deployed in digital twins so that they can benefit from continuous learning as they process live data. Now, streaming analytics applications can evaluate ML responses for live telemetry, flag invalid classifications, and create supplemental training data that feeds the retraining process. The platform periodically gathers training data generated by the population of digital twins, retrains the ML algorithm, and then redeploys it without interrupting live operations. Continuous retraining ensures that streaming analytics using ML maintains its effectiveness and avoids model drift over time.

Also Read: AiThority Interview with Yuval Fernbach, VP and CTO of MLOps at JFrog

How can digital twin applications enable better real-time monitoring and autonomy?

Our world has become increasingly dependent on uninterrupted operations from large, complex systems that run key infrastructures, like smart cities, security systems, transportation and logistics networks, telecommunications, military command and control, and many others. These systems require continuous, real-time monitoring to detect emerging issues and respond in a timely manner. They generate vast amounts of streaming data from thousands of data sources, all of which must be assessed in seconds.

To address this challenge, ScaleOut Software developed a fast, highly scalable streaming analytics platform that leverages the digital twin concept to track individual data sources and introspect on streaming data in real time. While digital twins were originally created to aid in designing complex new devices, like jet engines and wind turbines, they are well suited for streaming analytics. They can store dynamic information that aids streaming analytics in evaluating incoming telemetry. We host them in memory (not on disk) on a scalable platform that can simultaneously track millions of data sources.

As a simple example of what a digital twin can do, imagine a digital twin of an 18-wheeler transporting cargo that detects unexpected lateral accelerations. Because it has contextual information about that specific truck, it can examine the driver’s history and schedule to see if the driver is likely fatigued and needs to be instructed to pull over. This kind of deep introspection requires immediate access to relevant contextual information, and this is just what digital twins provide.

Tell us how AI is impacting and influencing real-time monitoring and operational intelligence.

The goal of real-time monitoring is to provide the most intelligent analysis possible continuously. This helps ensure that operations managers quickly identify emerging issues and never miss them due to errors or fatigue. By boosting situational awareness, real-time monitoring enables managers to evaluate problems and respond to them effectively.

AI techniques, including ML and gen AI, take real-time monitoring to the next level. ML algorithms can find anomalies in telemetry that are very difficult, if not impossible, for hand-written algorithms to detect. Gen AI can intelligently monitor analytics results the way a manager might and, in so doing, act as an assistant that never gets tired. It also can let managers use natural language to simplify the task of generating visualizations and queries that otherwise would require relatively arcane specifications.

What are you most excited about when it comes to the future of AI?

In the context of real-time monitoring, we are excited that as AI gets better, it can add more intelligence in assisting operations managers as they track issues in critically important live systems. As with airline pilots, managers of live systems must avoid mistakes that cause them to miss important issues or not grasp their extent and severity. The longer the delay in creating an effective response, the more difficult the task often becomes. AI techniques can mitigate this problem and do an increasingly good job as AI technology matures.

Also Read: AiThority Interview with Nadav Eiron, Senior Vice President of Cloud Engineering at Crusoe

A few daily best practices you’d share with developers who are building with AI today?

While our experience with AI is growing, we have learned the importance of careful testing to ensure that gen AI always provides useful responses to prompts. Our gen AI features need to remain accurate, adaptive, and aligned with real-world conditions, preventing outdated or speculative data from influencing decision-making. For example, to prevent AI hallucinations — situations in which gen AI produces incorrect or misleading insights — we ensure that responses are factually based on real-time digital twin data and constrain them using structured data outputs.

It is not practical today to have gen AI process the vast amounts of data collected and analyzed by a large-scale streaming analytics platform like ours. To avoid this, our platform aggregates results so that gen AI can evaluate a much smaller volume of data that nevertheless represents the dynamic condition of a large system. We have learned that we need to test these features at scale to make sure that we do not overwhelm the Gen AI system.

[To share your insights with us as part of editorial or sponsored content, please write to psen@itechseries.com]