I’ve worked in infrastructure software through my career, with a good chunk of that in distributed infrastructure. Believe it or not, I hold a patent in software-based edge routers from over 15 years ago, when the “edge” was just the networking edge. When edge computing came about in its present form, I could see its usefulness in a distributed, connected world. We were already working in AI and were intrigued about ways to apply AI efficiently for the edge.
What is it that you & your company are uniquely bringing to the edge market?
At Solecular, we’ve come up with a patent-pending distributed deep learning architecture, Maxwell. Various flavors of time-series data occur in various domains including IT, energy, industrial applications, and retail. When I say time-series, you can include sensor data as well as things like video feeds from surveillance, traffic flow cameras, and future media-rich time series. The data needs to be converted to semantic knowledge and acted upon, quickly and efficiently. The idea that massive amounts of edge data will be analyzed hundreds of miles away is already outdated. Also, the idea that there will be pre-trained models that know how to interpret sensor data from their surroundings in all cases starts to break down very quickly.
Our approach is to create deep learning models at the edge — our patent-pending method significantly decreases the compute needed to do that; turning IoT gateways into ML training engines. Our models are significantly more compact than the raw data as well — we can store them 100x to 1000x more efficiently than the data they represent. Moreover, models allow forecasting, backcasting and other analytics given they semantically represent the data itself. So the compact models can flow across geographically distributed networks towards applications that aggregate them to make predictions. Maxwell is evolving into a globally distributed brain that can make sense of sensor data from thousands of sources and connect that to business outcomes and automated actions.
Tell us more about the company!
We started in 2017, focusing on applying AI to data center power. While we got some interesting use cases in that vertical, we saw the massive expected growth in edge computing and broadened our charter. We were able to extend Maxwell deep learning, which we’d built for data center time series, to various types of data that occur in operational technology. My co-founder and our COO, Doug Cheline, is a Telco business veteran, recently having led organizations in the CDN/Video streaming world. Our third musketeer, Brian Atwood, is a customer success superstar, one of the early guys at Data Domain (now a division of Dell/EMC following a $2.4Bn acquisition).
We got some early traction in the data center space along with some angel funding. On the edge side, we’ve been able to attract the attention of much bigger household-name telcos. This year, despite the pandemic, we’ve also started recognizing revenue in the edge space. One customer is using our AI pipeline for predictive routing and tracking in their wireless edge network product. Another is testing out our pipeline for their decentralized IT platform.
How do you see the edge market developing over the next few years?
I see a future in which the edge, the core, and the layers of compute in-between will be one seamless distributed computing fabric. The inevitable challenge becomes one of creating a distributed application fabric and a service fabric that drape over this substrate. What Maxwell brings to this world is to provide a distributed intelligence service.