On his way out the door in the morning, Jim Jernigan sometimes jokes that he’s “going out to save the world.” The claim sounds far-fetched from someone who works in IT, but there’s a surprising amount of truth to it.
His team designs, builds and maintains the powerful servers that have helped Microsoft researchers create revolutionary technologies from Kinect to Skype Translator, as well as supporting scientific breakthroughs in things like HIV vaccine design and pinpointing causes of disease.
“I’m certainly not one of the smart folks who come up with those ideas, but we help run the platforms that help make that research possible,” says Jernigan, a senior research and development systems engineer. “I get to work on projects that are actually making a difference in the world.”
Along with having such an impact, Jernigan says he thrives on the opportunities his job gives him to travel to other countries, work with “the best of the best” people and have the kind of work-life balance many IT pros can only imagine.
He and his colleagues on the Microsoft Technology and Research engineering team architected and built a GPU cluster — a massive block of graphics processing units that takes up a large section inside one of the team’s labs in Redmond, Washington.
It does the ultra-speedy math calculations needed for machine learning, which plays a big part of many cutting-edge technology applications from predicting someone’s cancer survivability rate to detecting malware on computer networks.
Machine learning is the foundation of speech recognition and the reason Cortana, Microsoft’s digital personal assistant, knows what you’re saying even though you’ve never trained her to understand your voice. It’s also the root of Skype Translator, which gives people across the world who speak different languages the remarkable ability to talk to each other in real time.
Frank Seide, whose team built the first Skype Translator prototype, says creating its speech recognizer required inputting more than 1 billion samples of sounds — and doing that hundreds of times. The GPU cluster Jernigan’s team built did computations in a couple of days that previously could have taken months.
“You really need to do a lot of computing… and that’s what this GPU server farm enabled us to do,” Seide explains. “It allowed us to run the huge amount of experiments for building the tools themselves, and then building the best possible speech recognizer for our customers.”
Jernigan and his team have built other clusters that support a host of ground-breaking Microsoft Technology and Research projects, including the development of a promising design for an HIV vaccine and a state-of-the-art algorithm for identifying the genetic causes of disease.
Much of the research is fundamentally statistical but not as complicated as one might think, says Carl Kadie, principal research software design engineer. If you want to test a possible association between a DNA location and a disease, for example, he says “there’s a straightforward way to solve that if you have enough computer power.”
“Jim’s team provides Microsoft Research with some of the best computational resources in the world,” Kadie says. “I can get results for relatively big problems in sometimes a half an hour.”
Jernigan also helped build a cluster that was critical in shipping the very first Xbox Kinect, a technology that involves machine learning for both capturing movement and recognizing voice commands.
He feels lucky for many aspects of his career. He lives five minutes from his office. He’s gotten to travel to faraway places around the globe, including India, Beijing and Stockholm. But the best part, he says, is that his job doesn’t take over his whole life.
Unlike many IT positions in most any industry, he’s not required to be on-call. His phone doesn’t ring at 2 a.m. He gets to work “really hard during the day” and leave his work behind at the end of it.
“Normally when you’re in an IT position and something breaks, you’re there until it’s fixed. That means if you have to pull an all-nighter, you pull an all-nighter,” he says. “For me, I can break something at 4:30 on a Tuesday, and at 5 o’clock, I can say, ‘Well, I’m going home.’ You come back in the morning and you fix it, and everybody’s happy.”
He’s also grateful not having to field a constant stream of requests from people who can’t access their email or get their printer to spit out an important document they need right away, as many tech support workers do. Jernigan’s projects are on a much larger scale.
“We can touch networking, security, server builds, software, high-performance things. The breadth of technology we get to deal with is fascinating,” he says. “And for geeks like us, it’s nice to be able to learn in all of those different areas.”
Jernigan came to Microsoft in December 2008. Before that, he ran his own IT consulting firm, but the economic downturn took a toll on the business. He started out at Microsoft as a contractor and was hired as a full-time employee in 2011.
Microsoft Technology and Research especially interested him because “it’s a forward-looking organization,” he says, “so we’re always dealing with the latest technologies.”
In one recent project, Jernigan and his colleagues built on the design of another Microsoft project to see how many input/output operations per second (IOPS), a key performance measure, they could get for a storage cluster. The earlier effort had topped 1 million IOPS; his team scaled up that design to reach 2 million.
The effort gave researchers amazing performance that cost little more than $100,000, says Jernigan, who values having the freedom to attempt something that is both a huge undertaking and an experiment.
“This is a project I would never have had an opportunity to work on in a small business or a more structured division in a large company,” he says.