SDS 605: Upskilling in Data Science and Machine Learning

This is episode number 605 with Kian Katanforoosh, CEO of Workera and lecturer at Stanford University. Today’s episode is brought to you by Pachyderm, the leader in data versioning and MLOps pipelines. 

Welcome to the Super Data Science Podcast. The most listened-to podcast in the data science industry. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. I’m your host, Jon Krohn. Thanks for joining me today. And now let’s make the complex simple.

Welcome back to the Super Data Science Podcast. We’ve got an eloquent and deeply knowledgeable guest for you today. The iconic deep learning instructor and ed tech entrepreneur, Kian Katanforoosh. Kian is CEO and co-founder of Workera, a Bay Area education technology company that has raised $21 million in venture capital to upskill workers with a particular early focus on upskilling technologists, like data scientists, software developers, and machine learning specialists. He’s also a lecturer of computer science at Stanford University. Specifically, he teaches the extremely popular CS230 deep learning course alongside professor Andrew Ng, one of the world’s best known data scientists. For his teaching eloquence, Kian was awarded Stanford’s highest teaching award.
He’s also a founding member of DeepLearning.AI, a platform through which he’s taught over three million students deep learning. That’s insane. And he holds a master in math and computer science from CentraleSupelec near Paris, as well as a master in management science and engineering from Stanford. By and large, today’s episode will appeal to any listener who’s keen to understand the latest in education technology, but there are parts here and there that will specifically appeal to practicing technologists like data scientists and software developers.

In this episode, Kian details what a skills intelligence platform is, four ways that machine learning drives his skills intelligence platform, what frameworks and software languages they selected for building their platform and why, and what Kian looks for in the data scientists and software engineers he hires. All right, you ready for this special episode? Let’s go. 

Kian, welcome to the Super Data Science Podcast. I’m stoked to have you here. Where in the world are you calling in from? 

Kian: 02:40

I am calling from San Francisco, the city of fog. Very good to be here. 

Jon: 02:46

Nice. Well, you’re managing to avoid a very hot summer, I guess, in a lot of the world. So maybe a good thing this time around. So you are the CEO of Workera, which has raised $21 million from prestigious venture capital firms, like Andrew Ng’s AI Fund, as well as other funds. And so Workera is a skills intelligence platform. Prior to being introduced to you, I had never heard of something like that. So please explain to us what a skills intelligence platform is as well as the, what you call precision upskilling that you do in the platform. 

Kian: 03:31

Yeah, that’s a good question to kick us off. And I would say that a skills intelligence platform is used by enterprises and their learners to understand their skills, to develop their skills, to mobilize their skills. And we essentially empower organization by giving them actionable skills data that can drive innovation. And once you have skills data, you can use it for hiring. You can use it for upskilling. You can use it for internal mobility. You can use it for mentorship. But the real idea is to get that objective pulse on your strength, your areas for improvement, et cetera, et cetera. 

Jon: 04:19

Cool. So if a listener is out there who is wondering if they have a particular technical proficiency, if they have some data analysis or data science, or machine learning skill, they could go to the Workera platform and do evaluations so they can see where their strongest, where they could use improvement. That’s kind of the idea behind Workera? 

Kian: 04:43

Yeah. Yeah. Yeah. Yeah. Usually you may be a lifelong learner, you take classes, but sometimes you feel the classes you take are not really suited for your level of proficiency. And usually you feel like what’s limiting you in your career is your ability to be self-aware. What am I good at? What I’m not so good at? How do I compare to the best in class engineers, analysts, scientists out there? Is there any benchmark I can compare myself to? Then that’s the gap that Workera would fill for you. We’re an objective psychometrically calibrated assessment that allow you to compare yourself on a standardized basis with best in class people. Understand at a granular level the skills you have and don’t have across a variety of technical topics, data, AI, cloud, DevOps, et cetera. 

Jon: 05:30

Nice. So is that, you just mentioned like data, AI, cloud, DevOps, is that kind of where the sweet spot is for Workera? It’s these kinds of technical roles that our listeners, data scientists, software engineers are really interested in, or is it broader than that? 

Kian: 05:46

It’s even broader than that. We aspire to one day be able to measure every skill so that we can provide rich feedback to learners and their organization. And today we’re very much focused on innovation. So anything that touches innovation is part of the short term roadmap. Next year we’re going to start launching a few soft skills as well- 

Jon: 06:09

Oh wow. That was going to be my- 

Kian: 06:09

That can be measured on the platform. 

Jon: 06:11

That was going to be my next question. 

Kian: 06:13

Yeah. Yeah. [inaudible 00:06:14] A lot of data scientists, machine learning engineers, software engineers are actually very interested in, what are the skills that I can acquire to become a manager and how good am I at those skills today? 

Jon: 06:26

Yeah. So I very frequently ask guests and I will even be asking you later on the show, what you look for in people that you hire or what the most desirable skills are in data scientists and software developers. And almost always the answers that people have are soft skills. So specifically, I think the one that comes up the most is communication. Particularly the ability to communicate technical concepts to non-technical people that you need to get buy-in from. So it could be executives in your organization who are going to be funding you. It could be investors, it could be clients. And so I can see how valuable having those soft skills will be when you have those in the platform later on. It’s cool. 

Kian: 07:08

Yes. Yes. Very important ones. 

Jon: 07:11

Nice. So prior to filming this episode, you and I were talking about how the product, how the Workera platform and the skills building roadmap within it was developed specifically with the user in mind. So it was developed to be ideal for the user experience. So what did you do to make that happen? What was the kind of mental framework that you used? 

Kian: 07:37

Yeah. The first thing is very few of us have the opportunity to be mentored. And I really wish that everybody had a mentor because a mentor can change your life. And then you start analyzing, how can we bring that to everyone? What does a mentor do with a mentee? And you realize that there is a phase where the mentor is trying to learn more about the mentee’s capabilities, because if they don’t know the mentee, they will not be able to mentor them. So that’s a, call it a diagnostic phase, it’s really an assessment. And then once they have a better sense of the mentee’s skill, they can ask them about their future goals, their aspirations, their dreams. And then now you have the beginning of the journey where I’m at and sort of the end of the journey where I want to go. 

And that’s where the mentor comes in, leveraging their experiences, maybe other people they’ve mentored, careers they’ve seen evolve in order to make the recommendation.
This is a good article for you. This is a good person you should know. This is a good internship you should go after. Or this is a good class at the university that I recommend for you. And what we noticed is that even a good mentor has only mentored a handful of people in their life. Sometimes tens of people if they’re really productive. But when we designed Workera we realized, we are testing people across the globe, across roles, across careers, across industries. And so we have the opportunity to build a mentor that is inherently more proficient than a human mentor if only we can start replicating some of the human aspects that a mentor and a mentee relation between humans has. And so we designed the product with that in mind. It’s evolving as we speak, but in the next few months, there’s a lot of that that will make it to the platform with insights that we actually hear mentors give their mentees and we’re bringing to the platform in a more automated fashion. 

Jon: 09:39

Super cool. So Workera is a digital mentor that, so not only can it evaluate skills, like we’ve talked about so far on the episode, but it just clicked for me, but of course it also needs to be recommending next steps to fill in the gaps. So that’s how this kind of digital mentor experience comes in. That’s really cool. 

Kian: 10:00

Exactly. Yeah, exactly. That’s a very important, and that’s what we call precision upscaling to your question earlier, where there’s so much content out there today. Like you teach on Udemy, there is 200,000 classes on Udemy. So how are you supposed to figure out what’s good for you is a huge question mark. And so Workera really sitting on top of this ocean of content clarifying for people what matches the most to your proficiency level. If there is granular skills that you need to develop, do you actually need to take a course or can you find a faster, quicker lesson, article, video in order to fill that gap? That’s also the work we do and that’s the work a mentor is supposed to do.

 

Jon: 10:43

Nice. Maybe the Workera platform will be recommending my course for people who need to brush up their linear algebra and calculus for machine learning skills. 

Kian: 10:51

Indeed, indeed, indeed. Yes. 

Jon: 10:54

So sweet. So speaking of machine learning, AI and machine learning plays a big role in the Workera platform itself. Can you give us one to two case studies of how machine learning powers the Workera platform? 

Kian: 11:12

Yes, yes, yes, yes. And it’s going to get technical, but that’s what we love. 

Jon: 11:17

It is. The audience loves that for sure. 

Kian: 11:21

So we build a company with data AI in mind from the ground up. So when you want to measure someone, you first need to have sort of an ontology. We call that a skill ontology. Imagine a large graph where each node is a skill and each skill is tied to a certain cognition. A skill could be, hey, what is PCA? Tell me about it. And that’s a recall level skill. It’s a remember level skill. But you could also ask someone, can you extract the principle components of these data metrics? And that’s an applied level skill. It’s very different in terms of cognition. You can also put a data set in front of someone quite open ended and ask them to analyze the situation, create something, synthesize information. That’s also different cognitive complexities. And so you have a graph where skills are tied to cognition and that’s the work of psychometricians really to establish that. Over the last three years at Workera we have evaluated a lot of learners around the world at the skill level.
And what I mean by skill is not machine learning, is a skill like on LinkedIn. 

Machine learning at Workera is tens, maybe hundreds of skills. It’s broken down into sub domains, into topics, into individual skill. And so this graph is really massive and very rich. So that’s the first piece of technology. How do you build such a skill ontology. But then it gets interesting when you think about the data that was collected on this graph, where you can start understanding the cross correlation between skills. Because if I know that you can do two times two equal four, I don’t need to ask you, my mentee two plus two. I can infer with 90% confidence, it’s a conditional probability that you would be able to do two plus two equal four.
And so you take that concept and you expand it to thousands of skills. Where we have now an engine that can, is a capture adaptive engine that can ask a question in order to maximize information over the entire graph. And after 10, 20, maybe 30 questions, we can already provide reach skills feedback across hundreds of skills to the user so that they know better themselves. And this is a very important skill technology that we call skill inference. So hopefully that makes sense. That’s only part of the story because- 

Jon: 13:53

Crystal clear. I got a quick question for you which is that, is that graph, this skills graph ontology, so we talked about the nodes being, they could be recall skills or they could be, I can’t remember the other kind that you described [inaudible 00:14:08]. 

Kian: 14:07

Apply, synthesize, create, analyze. Yeah. 

Jon: 14:11

So are the edges between nodes, are those manually curated or data driven? 

Kian: 14:18

It’s a very, very good question. You have many ways to build those edges between nodes. You can, if you think about it, you can build it based on conditional probabilities. So let’s say we have evaluated many people across two skills and we know which ones they have and which one they don’t have. And so there is a correlation that we can make between having a skill and having another skill. Certain skills are predictive of other skills or not predictive of other skills. Certain skills appear in someone’s skill sets. There’s a lot of cross correlation in there. That can be used.
On top of that you can use natural language processing in order to establish how close two skills may be. There’s a lot of information online that has been written by humans and that can be used to determine if skills are close to each other or not close. So for example, if you look on Wikipedia, addition and subtraction are mentioned commonly around each other. And so inherently, those skills are going to be close to each other. And you have also the possibility to add manual tagging. And that’s part of what we do at Workera as well. There’s a ton of different methods to build that graph. 

Jon: 15:32

Awesome. That was a great answer. This episode of Super Data Science is brought to you by Pachyderm. Pachyderm enables data engineering teams to automate complex pipelines with sophisticated data transformations across any type of data. Their unique approach provides parallelized processing of multi-stage, language agnostic pipelines with data versioning and data lineage tracking. Pachyderm delivers the ultimate CI/CD engine for data. Learn more at pachyderm.com. That’s P-A-C-H-Y-D-E-R-M.com, like the elephant. All right. Now back to our show.
So yeah, so in short, you can use data to drive the graphs, but then you could also potentially override in some situations or add in manual tags in some certain situations. And that makes a lot of sense to me because the skill ontology, the skill universe is constantly shifting. Things like addition and subtraction, those are going to be relatively stable, maybe for centuries, but some other things, new approaches come in. And so by building your graph in this data driven way, it can automatically adapt. You don’t need to be having huge teams of people looking out for new skills and figuring out how they should be slotted into the ontology manually. So it makes a lot of sense to me. 

Kian: 16:54

For sure. For sure. 

Jon: 16:57

Nice. So that’s really cool. Is that kind of the main way that machine learning is used in the Workera platform or is there anything else as well? 

Kian: 17:05

That’s only one of the ways. So that’s what we call skill inference. Now, let’s assume you have a skill inference engine and you’re very good at it. So you’re very good at measuring people. There’s also an entire part around psychometrics testing that I did not talk about, but I can talk about now. Imagine you’re asking a question to someone. This could be a very good question, but it’s usually a very bad question because it’s very easy to write a quiz and very hard to write psychometrically sound assessments. So for example, when you answer a quiz on a course, usually it’s more of an engagement feature, more than an actual assessment. So at Workera, we need to make sure that what we measure is what was intended to be measured. That’s called validity. And there’s a lot of statistics behind the scenes where you analyze the answers of thousands of people to understand, how can we calibrate the difficulty of a given question, the discrimination rate of a given question? Is there any bias? Are the distractors the right one?
Does this question give away the answer to another question? 

So there’s a ton of statistics around that as well. And that’s what we build in an automated fashion at Workera. So that’s the second piece of technology and there’s two more. Where it gets really interesting is once you have someone’s skills profile on the ontology that we’re talking about, so you have a user, you understand their skills across thousands of nodes in the graph. Then you can start really helping them. What does it mean to help them?
It means you can start recommending the right content to them, for example. And that’s what we call precision upskilling. Is our ability to understand the ocean of content and make it easy for our users to build the skills in an effective manner. And really what the user wants is not to take a 200 hour course if only 10 hour of the course is relevant to them. 

Jon: 19:00

Of course. Yeah. 

Kian: 19:01

They want research papers if they’re proficient, but if they’re beginners, they don’t want research papers. So you have a lot of contextual information to use to build an effective recommender system. That’s part of what we do and then it’s … Yeah. 

Jon: 19:14

It is so cool. I love that. I might have to try Workera. This sounds amazing. 

Kian: 19:22

And the other piece that is very interesting is really once you have that skills data, you can do a lot at the organization level. And this is usually what enterprises do. It can help you put people on the right projects. It can help you identify mentors internally, or really promote people to be mentors for a group and match them with actual mentees that have complimentary skill sets to them. It can also be used in a hiring fashion because when you understand your skills gap internally at the org level, then you can start configuring the Workera assessments for your hiring pipelines in order to make sure that when you hire a new data professional, they are bringing something new. They’re meeting their rigor or the benchmarks that you have established internally. And so there is a ton of technology that goes in all those talent strategies that go beyond education and upskilling. 

Jon: 20:22

So cool. That is awesome. So in addition to being a digital mentor, the platform can also use machine learning to facilitate real life mentors. 

Kian: 20:34

Yeah. So it’s a mentor for the user, but it’s also a mentor for the organization, if you will. 

Jon: 20:40

Yeah. Nice. Well, that all sounds really good. My enthusiasm for this, I’m not putting this on for the listener, this sounds like an amazing tool that I would love to be using. So Kian, now that we’ve talked about what the Workera platform is and how AIML drives the platform, I’d love to hear a bit about what it’s like for you personally, day to day as the CEO of a venture capital backed AI platform like this. There’s probably a lot of listeners out there who dream of being in your shoes. What is it really like day to day? 

Kian: 21:16

It’s great. I love the job. More precisely, I try to be precise. About half of my time is spent on product delivery. And product delivery at Workera encompasses not only product or engineering, but product engineering, machine learning and psychometrics. And we have a very interesting org structure. Squads that are each driving a certain product outcome. The squads are, when you’re part of a squad, you have a product manager, you have a product designer, you have someone that usually knows AI and you have engineers as well. And so it’s a very versatile squad and you belong to that team driving that outcome, and so you can be focused on being independent.

 

And then these squads are led by our head of products and our head of engineering, and our head of AI, and our head of psychometrics. And I work very closely with them to make sure that we continually develop innovative solution that are useful.
So that’s about half of my time. 20% of my time is spent talking to users and customers to understand their pain points, help them unlock the product’s value to the fullest. 20% of my time is also spent with the team internally. So that maybe managing, that maybe helping unlocking people, aligning people on the company vision and strategy, and then maybe the last 10% is miscellaneous tasks. For example, keeping the board of directors up to date with the progress of the company. From time to time fundraising. Talking to investors. Sometimes hiring. Yeah, I would say that that fills up probably more than 10% actually. 

Jon: 23:13

Right. Yeah. It does sound like that would take up more than 10%, but I love that. So 50% of time on product delivery, 20% on talking to users, understanding pain points. So important, especially for the CEO to be doing that. I think that that is ultimately what drives your capability in that big 50% of product delivery time. I personally get a lot of the best product ideas for my platform from users directly. 

Kian: 23:40

Yeah. 

Jon: 23:42

20% managing and then 10% plus on miscellaneous tasks. Like dealing with the board, hiring, fundraising. Very cool. That was a really concise and clear description of what it’s like day to day. I don’t think I’ve ever had, I’ve asked that question before, I don’t think anybody’s ever broken it down into sections like that, into percentages. So how did you guys decide, it’s clear that you’re involved very closely, very intimately with engineering and machine learning, and product. So when you were setting up Workera, how did you make the particular framework or software language decisions that you made? Those end up being critical as a platform grows, so how did you make your decisions and what decisions did you make? 

Kian: 24:36

Yeah. That’s a very interesting question. So first thing to know is Workera is a remote company. Today we’re about 60 people and we’re split across 20 countries. It’s really distributed. The way you run a remote company is very different from the way you run a company in the office and to be effective, you need to bring certain levels of structure. And so at Workera we use Shape Up, [inaudible 00:25:09] of Shape Up, which is Basecamp’s model and it works very well for us. So one thing that we do is, just for people who are not familiar with Shape Up, is instead of saying, we have this roadmap, how long is it going to take us to build this feature or that feature? 

We flip the problem and we say, we have eight weeks. That’s our company’s heartbeat and what can we deliver in eight weeks?
It’s actually six weeks plus two weeks of cool down, but essentially, what can we deliver in this timeframe? So we flip the problem. And this makes it much more structured. It forces our team to be very lean and very iterative, which we love. It’s also good to keep the ropes tight when you’re remote and you don’t meet everyone on a daily basis. We do meet regularly in events. And so the reason I’m mentioning all these context is because it influences the way we build our products and the tech stack that we use. We started Workera with Python majorly because Python is an easy start. But one thing that we observed with Python is you can do so much with Python and it’s so high level that you have the flexibility to make mistakes as well. It doesn’t push you to think about the system at the ground level. 

So more recently, we keep using Python for certain things, but we also add it to our stack Elixir, which is a dynamic functional programming language. It has a very modern syntax. It takes more time to get used to Elixir, but the reason we like it is because it allow us to cut certain communication lines. And for a product data AI organization to be effective, you need to empower each engineer or scientist to do as much as they can. And with Elixir, one engineer can actually do so much. And so there is less back and forth because people can be full owner of a certain slice of the product. And this has been a good thing for us. And we are already seeing an increase in productivity for engineers who have embraced Elixir. 

Jon: 27:18

Wow. So there are a number of popular functional programming languages out there as options for you to work with. How did you choose Elixir specifically? That probably isn’t one of the top five most popular functional programming languages. 

Kian: 27:33

Yeah. I think it’s very trending. We also find that engineers who know Elixir are usually really good, honestly, and it’s a programming language that a certain type of people really like. Also, we had our head of engineering who spearheaded a lot of these decisions and I learn a lot from, who was very bought into the way Elixir deals with concurrency, the way it allows us to remove certain communication lines and the way it thinks about data types. And beyond that, we also had a few experts because when you do a tech stack change, it’s very important to have champions that can champion this change and actually educate the rest of the team to become proficient at a new programming framework. So this was also another tailwind for us. 

Jon: 28:27

Is Elixir one of the skills covered in the Workera platform?

 

Kian: 28:31

Not yet. Coming. 

Jon: 28:34

I bet Python is though. Yeah? 

Kian: 28:36

Yeah. Python is, TensorFlow, PyTorch, a lot of the packages on top of Python that are commonly used in data science.

 

Jon: 28:42

Nice. Cool. Yeah. So for now you need champions, Elixir champions, but maybe in the future you wouldn’t. You could just have people doing skills tests and taking the right courses online and upskilling themselves. Super cool. 

Kian: 28:54

Exactly. 

Jon: 28:56

Nice. Do you end up using, you just mentioned machine learning libraries like TensorFlow, PyTorch. Do you use those kinds of automatic differentiation libraries as well? 

Kian: 29:07

Yeah. Not as much as I’d like to today anymore, but the fun thing actually is I was using TensorFlow many, many years ago when there was only graph execution, eager execution didn’t even exist. And I’m realizing when I look at now the documentation that the tool has changed so much. And I think this is a reflection of how fast the AI space as a job category is evolving and how important it is to always remain up to date if you want to be effective and productive. And unfortunately, I haven’t been as productive over the last year in TensorFlow, but I’ve caught other skills as well in the meantime. 

Jon: 29:48

Yeah. So PyTorch forced TensorFlow to become much easier for eager execution with a TensorFlow 2.0 release. But that was actually kind of annoying for me as somebody who was teaching deep learning because all of a sudden I didn’t have nearly as much that I had to teach students.

Kian: 30:06

For sure. 

Jon: 30:06

No, I’m joking. But it was actually, it’s really great because it meant that in the same, say in a five weekend course, instead of having to spend a weekend or two just on how to set up your graphs and execute them effectively, I could then have just one weekend spent doing the basics. And then that would free up another weekend for getting into really cool stuff like generative adversarial networks or deeper reinforcement learning that there otherwise might not have been time for. So yeah, as you say, it is so cool this industry that we work in, I think I say this on air all the time. But it’s so exciting to be working in an industry where year over year things can completely change in a way that makes your life as a data scientist or as a machine learning engineer, or as a software developer, easier and easier. And it sounds like a Elixir, which I have not looked into very much before, is another one of these tools that is making development easier across the board. 

Kian: 31:02

Yeah. Yeah. Yeah. And to clarify, from a software engineering perspective we’re using Elixir, but for our AI practice, we’re still using Python naturally given the amount of frameworks and packages that run on top of it. 

Jon: 31:17

Yeah. That makes perfect sense. I could imagine a scenario, and I don’t need you to go into the details, but I could imagine there could be scenarios where data science teams are using Python to develop models, but then in your production systems, you could be taking the model weights out of TensorFlow or PyTorch, or Scikit-learn, or whatever, putting those model weights into an Elixir production system where things can run more efficiently. 

Kian: 31:45

Correct. 

Jon: 31:46

Cool. So it sounds like an amazing company to work for as well, especially for people that are looking for remote roles and especially for those listeners out there that already know Elixir, they might be rushing to the careers page at Workera right now. So what do you look for in the data scientists or the machine learning specialists that you hire, Kian? 

Kian: 32:10

I look for a mix of hard and soft skills. So here is how an interview would go. I would start by trying to understand the logic behind the person’s career decisions. So I don’t care at first about the projects they’ve worked on, but more, why did you take that job? Why did you leave that job? What happened? What is your mindset behind those decisions? And that would help me better understand the person. And then I will ask them about their most significant achievements. And the reason I do that is because I want them to choose a project that they know very well and that they should be able to explain into details. And then we dig into it together. 

And what it allows me to do is to benchmark their best work.
If I’m not impressed, then it’s not a good sign. I will try to gauge how much complexity there was, how many resources they had, how long it took and really benchmark it against projects that I’ve done myself. What I’ve seen my students do at Stanford. And this part gets very technical in general for the data science and machine learning, or software roles. So that’s the second part. And then I’ll focus on self-awareness and why they want to join us. So I may try to dig into what they believe are their strength, areas for improvement. And then if they have areas for improvement, I would expect them to tell me how they’re doing in order to cover them up or to upskill. And so I would expect to see a certain amount of upskilling. 

Jon: 33:45

That’s a funny thing to have in a Workera interview. 

Kian: 33:47

Yeah. The best answer is, yeah. Here is my Workera skills profile. You have my [inaudible 00:33:52]. 

Jon: 33:53

Exactly. That’s what I was thinking. So perfect. So yeah, you look for a mix of hard and soft skills. So you look for the mindset behind their career decisions. Then you dig into their most significant achievement in a technical way, and then you get a sense of their self-awareness by making sure that they’re interested in the company. And the way that you do that is making sure that they already have their Workera skills profile. [inaudible 00:34:19] 

Kian: 34:21

Yeah. We also use, for technical roles we use the Workera product actually to hire. So they go through the assessments. 

Jon: 34:32

Makes perfect sense. You’ve got to eat your own dog food, as they say in the tech industry. 

Kian: 34:37

Mm-hmm. 

Jon: 34:38

So you’ve talked about mentors a lot, we’ve talked about mentors a lot in this episode. And in your response to my preceding question, you mentioned how you teach at Stanford. And that’s something that we haven’t actually dug into that much yet so far in the episode. So you teach an extremely popular course, one of the most popular courses on the planet, CS230, which is a Stanford deep learning class. And you teach that alongside one of the world’s best known data scientists, Andrew Ng, and you are very highly recognized for that teaching. So you won the Walter J. Gores Award teaching award at Stanford, which is the highest distinction that you can get for teaching at Stanford. 

So clearly you’re an outstanding instructor. That probably wouldn’t be surprising to listeners to hear because you’ve been extremely eloquent and concise this entire episode. So it’s been a delight to have you on as a guest so far. So working with people like Andrew Ng, both as somebody that you teach with, somebody that you’ve been involved with previous startups with like DeepLearning.ai and somebody that through his AI fund is also an investor in Workera. Clearly you have a really enviable mentor right there in Andrew Ng. And then you also have other really well-known mentors like Dan Boneh. And so what’s it like being able to have your own mentors like this? Was this part of your inspiration for thinking up the idea of having a digital mentor? 

Kian: 36:19

Maybe, I don’t know. And that would be subconscious. But no, I truly believe that a mentorship can change people’s lives because it changed mine. And sometimes I genuinely think that I have a lot of friends that had so much potential but never were at the right place at the right moment or did not have a mentor to guide them. And on the other hand, I see people that probably had less potential or less hunger for work that ended up doing extremely well because they had a mentor. So that’s an important topic for me personally. When I arrived at Stanford, Dan Boneh was the first person really to give me an opportunity sponsoring my studies. I worked with him on the first cryptocurrency class at Stanford. Learned a lot from him. I ended up becoming the head teaching assistant for his cryptographic course, and that was a great experience.
I love cryptographic stuff from a science perspective and technical perspective. 

And then at some point I met Andrew Ng. I was obviously looking up to him having known him, having been one of his students before online, but now becoming his student formally at Stanford. He was one of the figures that I was hoping to work with and learn from as an educational entrepreneur. And I started working with him at Stanford. At some point, after working for some time in academia together, he has this plan to launch DeepLearning.AI and he knows my passion for education and my previous startup in education as well, and asked me to come help him.
And I was fortunate to be maybe in the right place, in the right moment with the right person. And through DeepLearning.AI, him and I have taught together AI to over three million people and growing. And I learned a lot through this experience. Later on I told him about the plan for Workera. Andrew’s actually the chairman of the board at Workera and also one of the investors in Workera. So we work together very closely even now. And he loves Workera, he loves the vision and he’s very supportive in that endeavor. So I’m very grateful to be able to work close to him. 

Jon: 38:57

Yeah. I can imagine. He’s an unbelievable mentor. So I’ve met him once. I was speaking at a conference earlier this year, the Insight Partners ScaleUp:AI conference, and Andrew Ng was one of the keynote speakers at that conference here in New York. And I was blown away by the amount of time that he would spend with anybody who came up to him. So on the first day of the conference, in the evening there was this cocktail reception. And of course he was completely surrounded all the time by different people who were hoping to get a moment to speed to him. Including me, I was one of these people waiting around. And while I was waiting around, I listened to him have very long in depth conversations with people, where he would very patiently, he had no sense of urgency. 

All these people are standing around waiting to speak to him, but he, one person at a time, look them right in the eyes and ask them really deep questions about their commercial strategy.
Ask them really deep questions about their tech, about their machine learning approaches. He would talk about other companies that they might be interested in or other tools they might be interested in. Papers they might like to read that could be helpful for what they’re developing. And I’ve never seen anything it. I’ve never seen anything like that level of patience and thoughtfulness with such a big crowd around in my entire life. Yeah. And then I got to experience it myself too. And I kind of, I wanted to, my main thing was, of all the things I could have asked him in the world. I tried to ask him to be a guest on this show and he was very polite. He gave me a card and I emailed him, and maybe someday I’ll get a reply. No, but he was also very upfront about, he is only one person and obviously lots of people want him to make appearances.
And so I made the best case that I could and maybe someday we will have Andrew Ng on the show, but we haven’t yet. 

But yeah, so he seems like just an unbelievable force to be aligned with. And so as chairman of your board, as an investor, as someone that you teach with, you’ve won the mentorship lottery or your hard work put you in a place where you were well positioned to win the mentorship lottery, making it, I guess not much of a lottery. But yeah, really great. And so we talked a little bit there about your other startup experiences. So let’s dig into that a little bit more. So you talked about how, even before meeting Andrew, you were deeply interested in education. It was something that you were already passionate about. 

So you were not only a founding member of DeepLearning.AI with Andrew, but while you were a graduate student in France, you co-founded another ed-tech company called Daskit.
So clearly there is a theme here. You are huge into ed-tech. You see a huge opportunity for technology to improve learning outcomes. And we’ve heard already in this episode and probably other listeners like me have been impressed by an upskilling platform like Workera. I wasn’t really exposed to this kind of idea before, but I can see how technology can make a huge difference to my education once I start using Workera. So how did you get started with this? How did you have your first motivation to be involved in ed-tech? How did this occur to you as something that you’d end up spending so much of your life working on? 

Kian: 42:58

I feel like it’s a passion I’ve had for a long time. Maybe most likely actually, because of my parents. My parents were students in Iran a long time ago during a revolution. And so they had to leave a country, start a new life from scratch. And then at some point you cannot focus on your studies anymore. You have to leave. And I always remember my father saying how he never managed to achieve his dream. He never managed to study. He wanted to become a scientist, but didn’t end up taking that route for life circumstances. And he always taught me that education can change your life in a good way or a bad way if you don’t have education. 

And so I always grew up in mind the fact that education is really what can level the playing field. What can unlock human’s potential.
And there’s one thing that our head of AI, Morteza says is that human capital is the most under utilized or inefficiently utilized resource in the world. There’s so many people that are hungry for work, that want to do more. They don’t know how to access opportunity. They don’t feel like they’re given a hand or an opportunity to do something, even if they want to do it. And so I always had, these questions have always been in my mind and today at Workera I get the chance to work on it. So it’s so exciting. And so I’m very happy. 

Jon: 44:38

Nice. That makes a lot of sense and it’s a great personal story. And I absolutely agree with you. I did a TEDx talk in the spring where part of my big thesis was that technology is accelerating at a faster and faster pace. And what allows that to happen before we, if artificial general intelligence, if AGI is possible in the future, then human brains might not be as essential at some point, if that theoretical point can be reached, the singularity. But up until then and possibly forever, we’re reliant on human ingenuity. And because there are more and more brains, there’s billions more brains every couple decades on the planet and all of those brains have more capacity to learn than ever before. So the amount of time that people, whether it’s in the rich world or not, need to spend, subsistence farming, it’s a very small proportion now.
Very few people are involved in agriculture or manufacturing. Those numbers are going down all the time. So people are more involved in cognitive roles. 

And also more and more people have access to the internet where through archive papers and Stack Overflow question responses and GitHub repositories, you have access to the most cutting edge things. So we were talking earlier about TensorFlow. So there’s huge change from TensorFlow one to TensorFlow 2.0, where all of a sudden you could execute things eagerly or it’s much, much easier to build your computational graphs. And somebody at Stanford University gets access to that at the same time as somebody in Sub-Saharan Africa or somebody in Southeast Asia. And so you can have somebody in Kenya or Indonesia who is learning about deep learning and coming up with great new ideas. And so I think I’ve covered this sufficiently now. There’s this unbelievable opportunity to be unlocking societal progress. And so I guess that was a super long [inaudible 00:46:55] to say that I agree with you [inaudible 00:46:58]. 

Kian: 46:57

But you see what’s interesting is at DeepLearning.AI we had users that were in Ethiopia, for example, and they have access, as you said, to the same courses that a Stanford student has. They actually, despite that they don’t feel equal to a Stanford student because what they’re missing is the mentorship. They’re telling us, the main difference between me and a student at Stanford is the proximity to the knowledge, the mentors. When you’re at Stanford, it’s very easy to know how far you are from the level of a Google engineer, because you’re surrounded by Google engineers. It’s very easy to understand how far you are from the level of a PhD in computer science, because you’re surrounded by them. But when you’re alone, somewhere in the world, despite having access to the content, you don’t get access to that mentorship. So part of what we do at Workera is skill mentorship, bring it to everyone. 

Jon: 47:52

Brilliant. Yeah. I’m glad that you were able to tie it back. That makes perfect sense to me. And I can see how valuable that would be, especially in those situations where people are relatively isolated to give people the confidence that, yeah, you’ve done CS230. You’ve had the same information as a Stanford student. And so now go off and apply it. Try to come up with your commercial idea in Ethiopia. Super cool. All right. So we’ve talked a bit about your background. You’ve done a couple of master’s degrees. So you did a master’s degree in France at CentraleSupelec. So that was a master’s in mathematics and computer science. And then you came over and did another master’s at Stanford in management science and engineering on the computer science track. And that’s where you had Andrew Ng as advisor.
So you haven’t done a PhD. You’ve done two masters, which is an incredible accomplishment. And simultaneously without a PhD, so I think a lot of people associate teaching in university with people who have PhDs. But yet here you are without one and you’ve won the highest teaching award that Stanford has. So do you have any thoughts for listeners on whether a PhD is useful to teaching or what circumstances somebody may or may not want to obtain one?

Kian: 49:23

Yeah. So I’d first say that a PhD is a title. It’s a certificate, it’s a diploma. And it’s one that has a lot of value, but we live in a world where these titles, certificates, prizes are worth less and less, honestly. What matters in a PhD- 

Jon: 49:49

Especially if you can do a Workera skills assessment- 

Kian: 49:51

Right. Especially- 

Jon: 49:53

And discover that you have better skills than the Stanford PhD student anyway. 

Kian: 49:58

Correct. But the value of a PhD is really the ability to connect with a mentor at the end of the day. And that mentor is going to teach you a lot. It’s also the ability to be in a lab and work with peers that are very talented. And so I feel like it really depends on the goals for your career. In data science we are lucky to be in a field where you can do a lot, actually, without having a PhD. There’s so many talented data scientists without a PhD. On the other hand, if you want to be an academic, then it’s important to learn from a mentor in an academic context and go through this paper submission, review process, rebuttal process. All of that is important to learn, how to read research papers. Those are things you learn in a PhD. And so if that’s your goal, then a PhD is very well suited for you from a learning perspective.
Now, when it comes to teaching. For a long time teaching, and even today, teaching and research are hand in hand. Meaning universities expect their professors to be both doing research and also doing teaching activities. I am against that. I actually think that those are two different skill sets. We all had professors that were extremely good researchers, but extremely bad teachers. And they also know it and they wish they didn’t teach. 

On the other hand, you have people who are great teachers, but are just not suited for academic research. So the job category has to change. It just has to change and it’s changing. Today if you go on Udemy, Coursera, et cetera, most people who teach actually don’t have a PhD. Coursera maybe there’s more portion of PhD, but generally speaking, anybody can go and teach on those platforms. And so when I was thinking about my career, I loved teaching.
I loved sharing knowledge that I had acquired, and I didn’t think I need a PhD to do so. And in fact, I was more interested in starting a company than doing academic research and that’s my personal career aspirations. And so I almost looked at the opportunity to be mentored with Andrew Ng, who’s has so much knowledge in entrepreneurship and in AI and be able to do what I like, which is start companies. And I actually believe I would not have been a good researcher anyway. So I probably saved a few years of unfortunate, desperate attempts to submit papers.

Jon: 52:28

Yeah. And now you get to have him as a mentor for what your dream was anyway. So I think that that’s a really good example. That you shouldn’t do a PhD for its own sake. That could be extremely stressful. If your goal is to be an ed-tech entrepreneur, then don’t do a PhD with Andrew Ng, get him to be the chairman of your board. It makes perfect sense. So while you were, I mean, I guess your whole life, so not only while you were a Stanford student, but prior to that, you’ve played a lot of football, what they call in the US soccer. And so we have coming up, pretty soon in a couple of months, we’re going to have the Football World Cup. So Kian, do you have any predictions for who’s going to win? 

Kian: 53:18

France. I hope my prediction is as, you remember Paul the Octopus a few years ago who was predicting games? 

Jon: 53:25

Yeah. 

Kian: 53:26

Yeah. I hope my prediction is as accurate. I don’t believe there’s been teams that won the World Cup twice in a row. I may be mistaken, maybe Uruguay did it a long time ago, but I hope France can do it this year. 

Jon: 53:40

Yeah. They’re certainly a strong side. They certainly are contenders to repeat. We’ll see how that goes. Good luck to them. [foreign language 00:53:50]

Kian: 53:52

One of the players and I were actually in the same high school, just as a fun fact.

 

Jon: 53:57

Oh, really? Who’s that? 

Kian: 53:59

N’Golo Kante and I were in the same high school. 

Jon: 54:01

No way. N’Golo Kante, he was big when Leicester City won the Premier League on a very low budget out of nowhere. Yeah. Very exciting football player. Amazing that you played with him. Cool. All right. So we’ve gotten everything from you. We’ve learned about Workera. We’ve learned about what it’s like to be the CEO of a VC backed AI platform. We’ve learned about great frameworks, software languages in this episode. We’ve had your football predictions. One last question for you is if you have a book recommendation for us Kian. 

Kian: 54:42

Yeah. I’m a science fiction fan. I love science fiction because it opens our mind to what the world could be. There’s a lot of technology in there. A couple of years ago I read The Three-Body Problem. That’s a trilogy. There’s three books. It’s from Cixin Liu, which is a Chinese author. But the book has been so successful that it’s been written in many other languages. And I highly recommend it because I loved it. I think it’s a great book. 

Jon: 55:17

Nice. Yeah. I’ve heard a lot about it and I look forward to the day when I can make time to read that. It sounds fascinating. And yeah, clearly, as I’ve already mentioned on the episode, you are an outstanding communicator and you have a lot of really valuable content to share with people. How can listeners follow your thoughts after the episode? 

Kian: 55:40

You can follow me on LinkedIn where I’m very active sharing thoughts and sharing updates on Workera or sometimes DeepLearning.AI. I’m sometimes active on Twitter as well. And you can also look at the people I follow on Twitter if you want to build your AI Twitter following. And then I’m also, you can also follow the Workera page where there’s a lot of updates on the technologies we build, if you are interested in how skills measurement or skills intelligence can change how people develop or mobilize their skills. 

Jon: 56:22

Nice. Great. We will be sure to include those links in the show notes. Kian, thank you so much for being on the program. It’s been an absolute delight and maybe in a couple of years, we can catch up again and hear how the Workera journey is coming along. 

Kian: 56:38

Yes, yes, yes, yes. Perfect. Thank you so much, Jon. 

Jon: 56:46

Kian was so warm and fun to record with. Filming this episode today was an absolute joy. In this episode, Kian fill us in on how machine learning models drive four key elements of the Workera skills intelligence platform, namely the skills inference engine, the statistical validity of psychometric assessments, precision upskill recommendations and mentor mentee matching. He talked about how valuable soft skills like communication and management are in the workplace for technical folks like data scientists. He talked about how he went with Python to build his platform initially. And while they still use Python for data science tasks like model development, Workera increasingly use the Elixir functional programming language for their production code. 

He also talked about how mentorship changes people’s lives, how education levels the playing field and how France are going to win the upcoming World Cup. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Kian’s social media profiles, as well as my own social media profiles at www.superdatascience.com/605.
That’s www.superdatascience.com/605. If you enjoyed this episode, I’d greatly appreciate it if you left a review on your favorite podcasting app or on the Super Data Science YouTube channel. I also encourage you to let me know your thoughts on this episode directly by adding me on LinkedIn or Twitter, and then tagging me in a post about it. Your feedback is invaluable for helping us shape future episodes of the show. 

Thanks to my colleagues at Nebula for supporting me while I create content like this Super Data Science Episode for you. And thanks of course to Ivana Zibert, Mario Pombo, Serg Masis, Sylvia Ogweng, and Kirill Eremenko on the Super Data Science team for managing, editing, researching, summarizing, and producing another superb episode for us today. Keep on rocking it out there folks, and I’m looking forward to enjoying another round of the Super Data Science Podcast with you very soon.