Technology
National Engineers Week: Building a Better Remote
To celebrate National Engineers week, we’re featuring stories about Comcast engineers engineers, their work and the invaluable contributions they make to Comcast and to their broader technology communities. This article features Vikrant Sagar, a lead engineer working on the X1 Voice Remote.
As an engineering Intern in 2013, Vikrant Sagar jumped at the opportunity to work on the development of the X1 Voice Remote, since it was an opportunity to apply his master’s studies in machine learning to a real-world technology.
At the time, the Voice Remote was in its infancy, available only as a smartphone app. A small engineering team was still honing the remote’s search capability to transform the viewing experience.
One major obstacle was that the app’s word recognition was too literal. While the phrase, “Watch Harry Potter,” brought up movie titles as you would expect, “Watch please Harry Potter” yielded no results.
It turns out the app’s grammar rules were too strict, and Vikrant with the help of Comcast researcher Robert Rubinoff applied machine learning techniques to relax them, so X1 could better understand a customer’s intent.
Following a successful internship, Vikrant was hired full-time, and he’s been working on the Voice Remote ever since, refining and improving technology that is now used by millions of X1 customers. Today, there are more than 12 million voice remotes in customers’ homes.
Even as the technology evolves, there are always new challenges to overcome. Although speech-to-text software has drastically improved, there are limitations with male vs. female voices and accents, sometimes resulting in spelling errors. So, instead of risking no search results at all, Vikrant and the team improved the search system’s autocorrect capability.
In other instances, they factor in relevancy to make sure the best results surface. One such example is “Chicago Fire,” which is both a TV show and a professional soccer team. Based on a customer's viewing and query history, the TV show might appear. But if the soccer team is playing at the time, the game broadcast shows up.
"It's our responsibility to decrease complexity from the user's perspective," Vikrant says, "even if that means increasing the complexity on our end."