“If two sentences have exactly the same meaning but no word in common, how do you make a machine understand that they mean the same thing?”
Down below rue la Fayette in the Paris office of Kayrros, Vincent Chabot leads the initiative of designing the “intermediate universe” of natural language processing (NLP), the intersection between human language and its meaning to a machine. How can meaning be translated into a quantitative representation that a computer understands? The challenge is training a machine to search through immense quantities of text-based internet data and group together various streams of text conveying the same message, regardless of whether the words used to describe it are the same.
“I discovered data science during an internship, and I loved it. So, I went to New York University (NYU). They have a data science department founded by Yann LeCun, who is one of the three fathers of deep learning. I spent two years there, and had the chance to work with a research team in NLP, which is that part of data science where you work on ‘text’ data, so you try to go from French, English or any other language and its semantics into a mathematical representation.”
NLP is emerging as a tool that will transcend pure language applications
Though originally stemming from the idea of automated translation, NLP is emerging as a tool that will transcend pure language applications. Kayrros uses NLP to monitor major disruptive events in realtime by searching the internet for indications of incidents like facility outages, unexpected shutdowns and LNG plant restarts. In today’s interconnected world, on-site observers hit Tweet before media outlets catch wind.
“What we do at Kayrros is super exciting. We started with energy, which is why I came here. Now, we’re adding to our capabilities by expanding our asset observation into new global markets. All of the new projects we’re working on are really cool, like 3D digital asset mapping. We use innovative technology, and our clients are telling us that we’re giving them information unlike anything they’ve had before.”
Kayrros teams work on a variety of different projects across different types of data streams, from optical and radar satellite imagery to AIS marine signals and global asset observation. The company culture promotes autonomy and experimentation in a fast-growing environment — something that drew Vincent to Kayrros in 2017.
“I wanted to join a very early-stage company with a really high growth potential. Today it’s both challenging and high-level in terms of tech and data science.
Everything is fast-paced at Kayrros — that’s what I tell to people who join. Recently, we had the case of one or two interns that really wanted an internship mission description. I can’t provide that, because new projects come in every day. You can ask me one month in advance, but if I tell you ‘you’re going to work on this,’ you’re going to come and switch projects. The two new interns were really nervous — I said don’t worry. Everything we do is exciting, it’s just a matter of ok, what are we working on tomorrow.”