It all started in 1976 with a curious and talented research student, Tore Risch, who then started focusing on databases and query processing after studies and research within the field of AI. His supervisor was Prof. Erik Sandewall at the University of Uppsala—a pioneer within Swedish AI research, and who was supervised in turn by the legendary computer scientist John McCarthy who coined the term artificial intelligence in 1956, and developed the Lisp programming language.
Tore completed his PhD in the database field in 1978, and after some highly inspiring years in Silicon Valley doing a post doc at IBM, working with financial expert systems at Syntelligence, being a visiting scholar at Stanford University, and working as a computer scientist at HP Labs, he was appointed professor in Engineering Databases at the University of Linköping, Sweden.
During all this time, Tore had been focusing on conventional databases which handle only one single type of data representation—the persistent table. In 2000, Tore came back to the University of Uppsala where he was appointed professor in Database Technology, and this time he got in contact with space physicists, in Uppsala and at Astron, the Dutch foundation for astronomy research, who struggled with a different kind of challenge regarding search and analysis of data.
Using radio telescopes, space physicists receive huge amounts of data in continuous streams, so large that it is not even possible to store all the data on any medium for a subsequent data analysis. Instead, the search and data analysis had to be made directly on moving data in streams, in real time.
This required a completely new approach which Tore started to investigate together with a PhD student, Erik Zeitler, who after completing his MSc degree in Engineering Physics had worked a few years in the internet industry, but then decided to go back to the academic world, looking for new challenges.
To be able to search directly on data streams, Erik and Tore started developing a new query language and processing techniques that could be used on any kind of data representations—vectors (series of values), matrices (2-dimensional vectors), tensors (ask a physicist about tensors and you will get a lecture), and more. And they also wanted the query language to contain all kinds of mathematical and search based operators, for any kind of search case.
They soon discovered that scalability became an issue, and that the CPU couldn’t easily cope with the task when the data stream became massive and the queries complex.
The solution was to parallelize the computing process on several CPUs but then distribution between the CPU cores became an issue, and so they had to develop a highly efficient distribution process.
An opportunity to measure their progress turned up when they found an academic challenge called the Linear Road Benchmark, aiming at comparing performance characteristics of Stream Data Management Systems.
Linear Road specifies a variable tolling system for a fictional urban highway system where tolls are determined based on changing factors such as congestion and accident proximity, and the task is to handle and analyze data sent out by each vehicle on the highway every 30 seconds.
In 2007, when Tore and Erik started their work, the world record was set at managing the data stream from 1.5 highways. Based on new ideas, they soon managed to build a system that could handle 0.5 highways on a laptop, and a little later 1.5 highways on a PC, which led to some excitement.
With eight cores on a discarded computer cluster, they reached eight highways, and after further theoretic and innovative advances they managed to handle 64 highways—a result that was published and presented at a conference. However, knowing that this was not the limit, they eventually got the opportunity to demonstrate their methods on a large computer cluster at Uppsala University and achieved an astonishing 512 highways, which became a nice conclusion of Erik’s PhD thesis in 2011.
At this point, Erik went back to the internet industry, working at the Swedish fintech firm Klarna, but he also kept talking with Tore about how their unique research results could be made useful in the industry. They both recognized that they would need some entrepreneurial experience, and a warm summer day in 2014, Erik went to have a talk with an old friend from his college studies, Jan Nilsson.
Jan, a serial entrepreneur with a background as a co-founder of several well-known and successful Swedish startups and internet companies such as Framfab, Bredbandsbolaget, Folkia, and HBO Nordics, immediately realized the significant potential of Erik’s and Tore’s brilliant research if turned into a product. The three of them continued to discuss the opportunities, and in 2015 they founded Stream Analyze.
Product development started with some further improved benchmarking results. A natural early use case was predictive maintenance, analyzing the data stream from sensors in vehicles, which had already been investigated by Tore and Erik through the EU funded project Smart Vortex during their research.
Initially, the computing was hosted in the cloud, in line with the major current AI and IoT trend, but the founders soon started to recognize the unique advantage of the small footprint of the software, since it would allow the technology to be installed on any device.
And this is where the edge-based strategy of Stream Analyze started to take shape, built around the main software tool, sa.engine, having its roots in the two main unique characteristics of Tore’s and Erik’s research—the capability of searching and analyzing data in large data streams in real time, and building database technology with an extremely small footprint.
Turning these unique characteristics into a mature product, ready for market and built on unprecedented and highly advanced technology, but yet easy to use—a little like Apple’s iPhone, but without locking its users into a closed ecosystem—has also been possible through another growing strength of Stream Analyze while more skilled people have joined the company—a deep and diversified knowledge in its field.
Or as Tore Risch puts it
“Together, we now have an unparalleled competence—essentially a complete department of computer science.”