Amazon Web Services (AWS) Senior Leader, AWS Deep Learning
UNC Kenan-Flagler Business School Master of Business Administration (MBA), Data Analytics, Entrepreneurship
Current Time 0:00
/
Duration Time -:-
Progress: NaN%

How did you get to where you are today? What is your story? What incidents and experiences shaped your career path?

Summarized By: Jeff Musk on Fri Dec 20 2019
It's a long one. I've been through different positions, different kinds of roles from academia for several years to cloud computing, to twinkl computing and finally machine learning diploma in particular. So let's start from pretty much from the beginning. I'm coming originally from Argentina. So I did my undergrad in Argentina in electrical engineer in particular. And then I did a PhD in the same program between my university in Argentina and France. So I lived there for a year and this is where I got more in touch with artificial intelligence. I was working on an artificial intentions lab in Toulouse. I work on a mental reality. Then I got a scholarship to come to the U. S. to work on a computer vision. So I did a second PhD in the U. S. working on computer vision. I worked on Autonomous Robots. Then I went to work between the street for the first time on I was a heavy user of modeling and simulation seeming like many researchers back in the day on, I decided to work for that company on. After I started, it became one of the numerical specialists for modeling in particular. At the same time, my work because of my numerical background and apply math background, I work on functions for a compiler that we were building. I saw the business grow very rapidly in very few years, So I decided to take a management position. After being the architect, principal, several tangled roles, I became a manager for the simulation for a couple of years. Then I got the offer to start a big data initiative at Akamai. And I spent four years doing that, starting, in particular, the streaming processing until I arrive at two and a half years ago, I started with the most. Most of my tenure was actually was with the computer vision, loved 126. So I was part of cloud cam, security cameras. For RIng cameras, I work on many things there from video streaming, computer vision, face recognition topics like that. In addition, to cloud engineering and large scale systems. Six months ago, more or less, I decided to go back to machine learning on deep learning in particular. I did that back in the day when I was a researcher. I'm leading now frameworks particularly MexNet, compiler technology plus things like the model on several other innovations that we have here at AWS. 

What are the responsibilities and decisions that you handle at work? Discuss weekly hours you spend in the office, for work travel, and working from home.

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
I am responsible for presenting the plans for the team, not waiting for a big part of the organization in particular. We're going through this right now. We go around a couple of times a year through that process. And I essentially work with program managers or other managers and researching what? What's the state of the art? What is going on? Another company's what is going on? Research on see what are the biggest needs? So then we're usually write some proposals. And I come up with some plan for the rest of the year. That's a big part of my work, especially is not continues but happens when it happens. This is a lot of work because it requires a ton of research. Working with scientists, working with other engineers, other engineering managers, principal engineers on, you know, the whole community, plus the Brookside. Then my daily days are more about tracking those projects. Make sure that things move. I have three monitors for two weeks soon to have under me, so I make sure that the projects are moving. They they are going on truck if they find any, any roadblock roadblock, some sometimes is. They have a plan, a given feature or given toll kid, for example, on with the team found out that it's more complex or they are using them. That's 1/3 party. So for the now doesn't offer what we expected. Or it's not the stables we wanted since like that. So when things like this happen, ah usually do at the dive on work with technical teams, toe finger, people toe. Make some decisions. For example, With move forward, we change the tool on, then communicate tow the rest off the leadership team. So the other part that they spend quite some time is working way signed. Signed on, for example, researching new ideas, doing prototypes or working with people doing prototypes. Both, um, it's just thinking about what were we doing? Probably one year down the road of six months down the road, the less parties managing people, monetary expectations, making sure that everybody has a career has ah, bill can be around career, not only not in a project come being be succesful his challenge also managing, observing whether someone is not performing as expected to others. Those cases, that's Ah, it's a big, big part off my my day and communicating, communicating, Clint results, Communicate them, tow my peers time dedicated to my management communicated. Tow my reports to make sure that you know what this happening is Flo sprung from one side to the other.

What tools (software programs, frameworks, models, algorithms, languages) do you use at work? Do you prefer certain tools more than the others? Why?

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
Well, depends on what we're doing. For example, for cloud, of course, we'll say AWS. We're part of that so everything is related to AWS. For example to train models, my team and I, we're different little bit because I'm more into experimenting things. So, for example, I'm a heavy user of Mxnet because we developed that. I also use PyTorch and TensorFlow very heavily. TensorFlow to work on conferences and PyTorch when whenever I had a chance. So I took a look very closely. Look at open source tools that are related to either machine learning and distributed training machine learning in general. But mostly lately has been more about deployment, not much about machine learning, of course. The typical ones, the basic ones like NumbPy and all these things which are assumed that we all know. Programming languages, Python is very popular, all across the whole team. And this changes depending on the company, on the position you have in the company. In my particular team, Python is very popular. But the big part of the team also program in c++ because we're right in optimization. So I have one thing that the right code in java so that they will, show FBI for deployment. So this is why I hire people who could show expertise. So then I mentioned algorithms are going means well, because given that we we build a friend or castigation Eric. So we go from the NLP greens, you know, a modest big models, for example, training birds or kind of that kind of morals. Um, we use distributed training organisms were stools like horrible the distribution or by P s, that little distribution? Um, the So we have, ah, huge majority off our dreams. We do many, many or many things on computer vision. So we work quite a bit on competition. Our rings

What things do you like about your job? Were there any pleasant surprises?

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
Every single position I've been in has been a challenge. The pleasant surprise is I take those teams who have a bad reputation. They're too young or junior to get started. And then they've started to deliver like everyone else. I've done this several times in my career. The other pleasant surprises were I had two teams, science, and engineering, who were not working very well together and now it's very hard to say which person belongs to which team because we became one. I spent a lot of time building teams and relationships. Of course, I have to know excellent performance, not so good performance and some very few low performers, but so that means the individuals also stand out in the team. But having this idea that we are not only one person, that we are a team and we're gonna achieve greatness only by becoming a team. This is something that I very proud of doing that through my career and repeatedly that this for many, many times.

What are the job titles of people you routinely work with inside and outside of your organization? What approaches do you find to be effective in working with them?

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
We have depending on the title, something comes to the background of this person on the mentality, so it's key to understand the way they think. So let's go from the top. I have a word with VP of, for example, the VP of 126 or AWS. So all the statistics they see of AWS. Very high-level people. The way we approach these people is by giving the meaning of what we're doing. Not the technical details. So we got to be very careful about overemphasizing technical details because people will be lost if If we don't give the meaning of why we're doing something. What impact, what benefit will have to the company, to the user acceptable to the business? So the other very different mentality is the scientist. We pursue some particular goals for the sake of the bus in science, even though it may not really mean more revenue or any practical application. What kind of freedom, the kind of word, what kind of results completely changes on the way we were gonna talk is by letting the person express and it's kind of an artist like the experience, let this person fly on, get ideas and sometimes fail and give him more time to recover and then start again. Sometimes you give this person, some time to failure on a project that probably not that meaningful, but it's for the sick of keeping this person engaged. The other way to manage people from different countries, for example, cloud people, people from operations that have very mindful of, or this mindset of, security or automation very different or be very predictable everything that we do. To keep these people engaged is very difficult because they need to feel that things won't change even though we're changing all the time. For different levels of engineers. So for developers, applied scientists. it's similar to a regular scientist that we have to give usually more freedom than software developers. Software developer sometimes writes beautiful code and the way I approach them is trying to balance the beauty with the reverend results to get things out. The other is my peers sometimes are not having the same background. On the contrary in many cases I didn't need to make an extra effort to explain, for example, building an engine, or building a high-performance computing software that other people understand the complexity and the time I take you to do something we call it. I almost forgot the business people, and by business, I mean the project managers, support architects, sales business developers. So depending on the problem with the language of the communication, the level of detail changes completely. And finally customers, professionals have started with that. I mostly interact with enterprise customers. But lately, I've been interacting quite a bit with the other good professors, researchers, other people that hold a PhD. So then it changes the way we are communicating. And the way it's more about creating partnerships, creating interactions and seeing the opportunity to collaborate, for example in different projects from different agencies. 

What major challenges do you face in your job and how do you handle them? Can you discuss a few accomplishments?

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
One is communication. Particularly, I had this feeling. It was similar when I started with big data many several years ago. It's not very well understood. Machine learning is still new. Especially deep learning still new. We don't have really very well established rules. No, we don't have enough education yet. It is coming. It is slowly building. So it didn't start something to communicate different levels on different people. The understanding changes a lot from one place to another. So it's not homogeneous at all. That's a challenge. I faced the same challenge when I was working on big data, streaming data, I did it before it was popular and it was hard to communicate in particular with top management of the company the meaning of why we need it, why we had to do it. Essentially, when someone works on innovation, that's probably one of the biggest challenge. The other big challenge is communicating because of this problem of broken communication, understanding the personalities, it comes to conflict, right? Sometimes I mentioned before we got two teams not really talking to each other. So on my accomplishments are I deliver things. That means I was effectively communicating for I came and I delivered a big project. We had more than 200 people working on that, 250 people at some point. We ended up pushing this across the finish line. Back when I was at MathWorks, we had to change the whole infrastructure of the compiler and to make that change, I was the architect for that. I had a ton of assistance. But in the end, the whole computer became adapted to my architecture in the future. So sometimes the challenges is we have to have a backbone and keep going on. Believe in what we do and get better, express better and communicate better. The latest accomplishment which is very new and this is what I'm very proud of cause I'm going through that Battle. It is the change in the relationship between the teams, we deliver things that nobody thought we could. So six months ago, when I took over this organization, nobody believed we could do this on. We surprised everyone, but not only with surprises because of the quality and we finished with our plan. We finished working with the other team or partner team in a way that nobody has done it before. One week ago, when I was at NeurIPS for one senior person off the company that one year ago, the leadership of the group thought that it was impossible. She told me that was really a miracle. So yes, these are the accomplishments, it can go off course. There can be ups and downs but right now it's going well. 

What are the recent developments in the field? How significant are these improvements over past work? What are their implications for future research & industry applications, if any?

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
The biggest one said I really like is the automation as everybody writes. Computer vision is becoming very well established, and we see many companies adopting that from autonomous driving or not even autonomous driving but assisted driving. Today we have very few cars that can claim that they are semi-autonomous. But we have many cars that can claim that they can detect a collision. They can detect whether someone is falling asleep on the wheel or is deviating from the lane. This is common today with several auto companies, even though they don't claim that they have autonomous cars, we see that they're using this technology is a computer vision and other things to help enhance the driving experience and make driving safer. This is one of the nice outcomes of these. The other part is, one year ago when Google presented BERT on the results in NLP, there has been a huge wave of change in speech, in particular, and communications and text analysis. We're going to see more applications as the model will get smaller on cheaper so as it is still expensive. But we're going to see more applications coming up. And I expect many healthy and good applications that can help in communications for people that have problems, difficulties speaking or hearing and that I dream to see this become a reality. That way we can help people, in addition, to build the business around. The thing that I see coming which is most futuristic in the coming two to three years is reinforcement learning. I see some very good applications in Reinforcement learning for things besides games. Today, be gaming is still dominating the application of Reinforcement learning especially the visual is good. So I think some of the applications like inventory control and distribution. Besides games, I think this is gonna be more mature into coming two to three years. So this is mostly what I see. And then the next step is how we make everything cheaper, more explainable, more secure and faster.

What was the hiring process like for your job? What were the roles of people who interviewed you? What kind of questions were asked?

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
The interview for Amazon was quite some time ago. I interviewed for a role like this. So the people there were mostly directors. The directors of software development or very senior managers. So I had one principal architect that interview me that was more about architecture, talking about architecture of the science. And the rest was about the management, about approaching management, people management, expectation management. So, what were the main outcomes for the work, achievements, what have I delivered under what conditions? Have I developed the people? Usually, I find every day my show like performance province with people of low performance, high performance, people leaving the company dissatisfied, hiring and attracting talent. It's very important to track very high-quality talent and make sure that we're evaluating talent in the right way. That was essentially most of the content of the interview. Process? usually, it will be one or two phone screens and then normally on-site. I've been through different interviews through my life and in some cases and companies asks for a presentation at the beginning of the day and making a presentation, describing not that they were detail because it's not expected for my role but expected some of the technical complexity of the project and how did I face or someone has faced the problems that happened with someone. Let's say the project I mentioned before had this pressure that had more than 200 people involved so there were a ton of technical challenges and they were inspected. How to deal with stress so and then after that presentation, if it's not a formal presentation, one person asks about that, then usually there are at least five interviews that gonna be about different parts of management. as I mentioned, expectations, people, project management and I would say stress management. So there is also a lot of that performance management. And then there is always a least one person asking questions about career expectations, things like that. So this is this is a very common process.

What qualities does your team look for while hiring? How does your team interview candidates?

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
I look for self-starters, people that gonna commit and they want to learn. People who will deliver results depending upon their role. Of course, in the case of applied scientists is different because I mentioned before I give more freedom. But let's say for a typical compiling engineer or for a developer, I would expect someone that shows that there is no small choke. Person gonna do whatever it takes to get things done with good quality. So, we all take shortcuts, but we never want to compromise quality with those shortcuts. So the process is simple. One or two phone screens usually calling in one of them on after we finish with the coding part. After the person passes the phone screen, we bring this person on-site and we do know at least three technical to behavioral interviews. Something three and three. I usually do both technical and behavioral. 

What are different entry-level jobs and subsequent job pathways that can lead students to a position such as yours?

Based on experience at: Senior Leader, AWS Deep Learning, Amazon Web Services (AWS)
Summarized By: Jeff Musk on Fri Dec 20 2019
Entry jobs? Applied Scientists. The shortest path to my current position would be to b e areally good applied scientist with a strong background in machine learning. but then add to that component, software development practices and become both, applied scientists that can really write code and develop and produce, build products. Also, as an engineer, that becomes more of a mathematician, which further has to code very well and then learn about machines. This is a longer path to my road as a single person. I would strongly recommend someone to really become a man should think about going to business school, get strong training in managing people, managing the expectations, managing projects. All this management as I mentioned before. So that's something I did after finishing my Ph.D., I did MBA. It's very important especially when we try to build high performing teams that I tried to do that. It's very important to be aware of the strengths and weaknesses of the people we manage, make them play on the restraints in particular. I usually take their strength, their different techniques. I take the strength approach, make sure that we put the best. The right person in the right task and let them grow. So there is the technical component and then the managerial component. It is very important that the person takes management as a career and be educated as a manager. Technical people believe that because we're very good technically speaking, we can become a manager. It's not the same. It's very different. We need to be educated like we learn math early in the career.

What were the responsibilities and decisions that you handled at work? Discuss weekly hours you spent in the office, for work travel, and working from home.

Based on experience at: Senior Engineering Manager, Akamai Technologies
Summarized By: Jeff Musk on Fri Dec 20 2019
When I was there a few years ago, I was working on big data. I was there as a senior manager. It was not that different from what I do now. We started this big data initiative and most of my responsibility was to build a good proposal. To get the authorization and the funding on the stuff and to start building this project. So back then, I worked with some product managers to build a good business position for the product on the reasons why we had invested in that. I also hired the team and tracked the progress for all the teams involved. So I was back then working more on architectural designs. So I got five architects working on that project, and I was leading the architects and putting pieces together. They would work addict in different parts of the system, so I kind of stitched them all together. I was also managing some 24*7 services. So we had operational component, that means, for instance, things were going down, data centers not responding and things like that. So I had to make sure that the operations were still up and the services were up and running for operations worldwide. Weekly hours? They haven't changed much. Usually, I am in the office from ten to seven. I start my day around six. Work at home at the beginning, and then when I come back, I usually work again. I don't travel much, which is good. In my previous role at Amazon, I was traveling in 90% my time, so pretty much I wasn't at home ever. My level is pretty flexible because we are expected to work a lot and expected to be pretty much available on the phone over weekends and every single day of the week. So that means we can choose when to be in when the office. So but my regular day, if nothing goes wrong, I will start at around six or seven and finish usually, I go to the office around 10. Have a very short commute by way. And come back around seven and I take a break. Probably and then work from eight to ten or eleven.

How did the program prepare you for your career? Think about faculty, resources, alumni, exposure & networking. What were the best parts?

Based on experience at: Master of Business Administration (MBA), Data Analytics, Entrepreneurship, UNC Kenan-Flagler Business School
Summarized By: Jeff Musk on Fri Dec 20 2019
The best part was interacting with people who approached the problems in a different way from me. So that is why I believe it's very important for technical people to go through that. As a technical person, I had a very structured way of approaching a problem and when we were doing homework or projects, the way we start putting the ideas together was very different. So that was very different. The part I enjoyed the most was the diversity. I didn't have diversity before in my career. Then the faculty was good. I wish we could have been under more pressure. I like to be under stress. So that was okay. and resources? We're fine I can't complain. I actually was giving my fieldwork two days ago about that. That I would like to be more involved. Bond. I have many people that I know who have a lot to give back to the University or their school. And I believe schools are wasting an opportunity by not engaging us. I talked several times, and there were several people in the leadership of the school that they were thinking we should give talks about what we do and in the end, noting concrete and I don't have time to push for a long time on that, I can only offer myself and my time. It's similar to the previous one. It's very important to keep people engaged. I think this is a very virtuous circle if we can talk to the students or students can talk to us and give the field book. Learn about careers and mistakes. In addition to the regular assignments, they have to do every week.