Four Montreal students take first place at HackHarvard
“HackHarvard was maybe my 10th hackathon,” said Nicolas MacBeth, a first-year software engineering student at Concordia. He and his friend Alex Shevchenko, also a first-year software engineering student, have decided to make a name for themselves and frequent as many hackathon competitions as they can. The pair have already participated in many hackathons over the last year, both together and separately. “I just went to one last weekend [called] BlocHacks, and I was a finalist at that,” said MacBeth.
Most notable of the pair’s achievements, along with their other teammates Jay Abi-Saad and Ajay Patal, two students from McGill, is their team’s first place ranking as ‘overall best’ in the HackHarvard Global 2018 competition on Oct. 19. According to MacBeth, while all hackathons are international competitions, “HackHarvard was probably the one that had the most people from different places than the United States.” The competition is sponsored by some of the largest transnational conglomerates in the tech industry. For example, Alibaba Cloud, a subsidiary of Alibaba Group, a multinational conglomerate specializing in e-commerce, retail, and Artificial Intelligence (AI) technology, as well as Zhejiang Lab, a Zhejiang provincial government sponsored institute whose research focuses on big data and cloud computing.
MacBeth said he and Shevchenko sifted through events on the ‘North American Hackathons’ section of the Major League Hacking (MLH) website, the official student hacking league that supports over 200 competitions around the world, according to their website. “We’ve gone to a couple hackathons, me and Alex together,” said MacBeth. “And we told ourselves ‘Why not? Let’s apply. [HackHarvard] is one of the biggest hackathons.’ […] So we applied for all the ones in the US. We both got into HackHarvard, and so we went.”
Essentially, MacBeth, Shevchenko, Abi-Saad, and Patal spent 36 hours conceptualizing, designing, and coding their program called sober.AI. The web application uses AI in tandem with visual data input to “increase accuracy and accessibility, and to reduce bias and cost of a normal field sobriety test,” according to the program’s description on Devpost. “I read a statistic somewhere that only a certain amount of police officers have been trained to be able to detect people [under the influence],” said MacBeth. “Drunk, they can test because they have [breathalyzers], but high, it’s kind of hard for people to test.”
MacBeth explained that the user-friendly web application could be helpful in a range of situations, from trying to convince an inebriated friend not to drive under the influence, to law enforcement officials conducting roadside testing in a way that reduces bias, to employees, who may have to prove sobriety for work, to do so non-invasively.
Sober.AI estimates the overall percentage of sobriety through a series of tests that are relayed via visual data—either a photo of an individual’s’ face or a video of the individual performing a task—that is inputted into two neural networks designed by the team of students.
“We wanted to recreate a field sobriety test in a way that would be as accurate as how police officers do it,” said MacBeth.
The first stage is an eye exam, where a picture of an individual is fed to the first neural network, which gives an estimation of sobriety based on the droopiness of the eye, any glassy haze, redness, and whether the pupils are dilated. The second stage is a dexterity test where individuals have to touch their finger to their nose, and the third is a balance test where people have to stand on one leg. “At the end, we compile the results and [sober.AI] gives a percentage of how inebriated we think the person is,” said MacBeth.
“Basically, what you want to do with AI is recreate how a human would think,” explained MacBeth. AI programs become increasingly more accurate and efficient as more referential data is inputted into the neural networks. “The hardest part was probably finding data,” explained MacBeth. “Because writing on the internet ‘pictures of people high’ or ‘red eyes’ and stuff like that is kind of a pain.” MacBeth said that he took to his social media pages to crowdsource photos of his friends and acquaintances who were high, which provided some more data. However, MacBeth said his team made a name for themselves at the hackathon when they started going from group to group, asking their competitors to stand on one leg, as if they were sober, then again after spinning around in a circle ten times. “That was how we made our data,” said MacBeth. “It was long and hard.”
Participating in such a prestigious competition and having sober.AI win ‘overall best’ left MacBeth and Shevchenko thirsty for more. “HackHarvard had a lot more weight to it. We were on the international level, and just having the chance of being accepted into HackHarvard within the six or seven hundred students in all of North America that were accepted, I felt like we actually needed to give it our all and try to win—to represent Concordia, to represent Montreal.”
MacBeth and Shevchenko have gone their separate ways in terms of competitions for the time being, however the pair’s collaborations are far from over. Both are planning to compete separately in ConUHacks IV at the end of January 2019, where MacBeth explained that they will team up with other software engineering students who have yet to compete in hackathons. “We’re gonna try to groom other people into becoming very good teammates,” said MacBeth.
The first-year software engineer concluded with some advice for fellow Concordia students. “For those in software engineering and even computer science: just go to hackathons,” advised MacBeth. “Even if you’re skilled, not skilled, want to learn, anything, you’re going to learn in those 24 hours, because you’re either gonna be with someone who knows, or you’re gonna learn on your own. Those are the skills you will use in the real world to bring any project to life.”
Feature photo courtesy of Nicolas Macbeth