My story is pretty long but will try to summarise as much as possible. First of all, where am I now, now, I'm a senior research scientist at Uber AI, which is the organization of uber that does both from the mental research in machine learning and Computervision and enforcement learning and at the same time does collaboration with different teams within the organization, the whole bigger organization of uber to apply machine learning and acknowledging Uber problems. I got here about three years ago, from an acquisition, I was working for a startup called Dramatic Intelligence in New York. We were between 10 to 15 people, mostly Ph. D's and the startup was doing again basic research in machine learning. We had an algorithm that was really useful for many things that uber wanted to have. Plus, they wanted to hire us to start an organization within the company. Before that, I actually worked at IBM Watson and I was doing NLP with deep learning and also work on several tasks, in particular, we were applying that technology to enterprise search. And even before then I did the pH. D., question answering the University of Bari, Italy, and during my Ph.D., I spent some time doing an internship in Barcelona. I loved it there. Like every city I go to Europe, there's someone, and I'm also someone in the U. S. There's someone from that team, so it was a really nice experience. I would say the most incidental thing is most likely is me joining dramatic intelligence startup because I was working for IBM, I was absolutely fine there. I was working with one of the members of the regional team. I was really doing well with them, it was an amazing experience. But this was a moment in time. I met with the people of dramatic intelligence with Gary Marcos and that was completely out of the blue. I just met them at the meetup and I had a really nice conversation with them. They suggested that I would interview with them. I did. And the team was amazing and I decided to join them. That was really unexpected and at the same time was really unexpected that we were acquired by uber, relatively a few months after. Although we were doing some amazing work, it was not that clear. Not all startups get acquired, let's put it this way.

As a research scientist, most of the responsibilities are to have both research and some applied project. So, for research, I'm responsible for starting and pursuing some specific research lately, mostly in dialogue. But we also work on graph learning or enforcement learning, some computer vision, and some language generation work. So my responsibility is to start this project and push them to completion and publish papers mostly on these products and then the other products that are more on the applied side, where my responsibility is usually to work together with a team of engineers, data science within the company and to provide them with the tools like machine Learning Technologies and NLP technologies that are useful for their specific task. And if those two tools do not exist to create new ones. So, for instance, one of the outcomes of these collaborations has been on Ludwig, which is an open-source project that Uber AI released last February and that's one of the main things that I have that I've worked on, which is a tool that allows people to train the plenty models and use them for obtaining predictions without having to write code and again was developed internally because that's something that's super useful for our customer teams.

There is a slight difference between the projects that are more research-oriented than the ones that are more production oriented. In general, for instance, most of the stuff that we do is in python and when there's some production stuff, we end up using Scala Spark java that kind of stuck and some other things used for, like distributed computing and similar are with most leniency but interfacing with python. The technology that we use, we're kind of agnostic. We respect to meet the frameworks again we use both TensorFlow, PyTorch but at the same time, we create our own tools. So, for instance, Uber AI has also released this other tool, it's a probabilistic programming language called Pyro, it's built on top of PyTorch and also the one I told you before Ludwig, which is built on top of TensorFlow. So depending on the specifics of the products, specific tooling that we want to build, we start from a base of different technology, we're not extremely committed to one or another.