Career Growth Advice from Parmida Beigi, AI Leader | Career Tips for Women in AI

2B Bolder Podcast – Episode 81
Featuring Parmida Beigi

Episode Title: #81 Parmida Beigi: Innovating the Future at Amazon

Host: Mary Killelea
Guest: Parmida Beigi



Mary Killelea (Host): Hi there. My name is Mary Killelea. Welcome to the To Be Bolder podcast, providing career insights for the next generation of women in business and tech. To Be Bolder was created out of my love for technology and marketing, my desire to bring together like-minded women, and my hope to be a great role model and source of inspiration for my two girls and other young women like you. Encouraging you guys to show up and to be bolder and to know that anything you guys dream of, it's totally possible. So, sit back, relax, and enjoy the conversation.

Thank you for tuning in. Today's guest is an incredibly smart woman who is on a mission to demystify artificial intelligence. She is a huge advocate for helping women succeed in their technical careers. Parmida Beigi is an Amazon senior research scientist. She shares a lifetime worth of experience and uses her skills to help others grow into machine learning career paths. Parmida's career has touched many facets of machine learning and data science, from her PhD research in computer vision and time series forecasting to her work in Alexa AI end-to-end systems. Today, Parmida pursues, among other things, speech recognition and natural language processing initiatives to help Amazon's Alexa customers. I am so thrilled to have you here. Thanks for joining.

Parmida Beigi (Guest): Thank you, Mary. Thank you for having me. I'm really glad to be here. By the way, you pronounced my name really nicely, like perfectly. Thank you for that.

Mary Killelea: Yes, of course. Okay, so I read, you know, in doing some of the research for the show, you know, you started out thinking you might become a doctor, but then through the schooling, you enjoyed and gravitated early on towards math and science and ended up earning a bachelor's degree in electrical and computer engineering, followed by a master's. And then you focused your research on comprehensive sensing, signal processing, image video processing, and then you went on to get your PhD in electrical and computer engineering, which is so, so impressive. Take us back and tell us about your career path and how you ended up where you are today at Amazon.

PB: Yeah, absolutely. So since I was a kid, I've been really into mathematics and computing, but I was also intrigued by medicine, possibly due to my exposure to medical discussions at home. But the more biology classes I took, the more I realized math and computing is in fact where I belong. So, I started engineering. One of the pivotal points in my career is the research I did at BC Cancer Research Center, where I touched on classical machine learning and statistics techniques that was back in 2012. And it's really interesting to me because that is exactly the year when the Harvard Business Review article on data scientists being the best job in the market was published. And it's also the year when machine learning and AI, specifically convolutional neural networks actually hyped up. And that was the proposal by Jeffrey Hinton from University of Toronto, Alex Net and the revolution in computer vision.

So, for my PhD research after that, I first started working with classical computer vision on time series forecasting methods, but I found myself increasingly drawn into machine learning and the conferences I was attending and journal and conference articles I was reviewing. So, I started learning and practicing machine learning more systematically and started employing machine learning in my PhD research as well. And upon my PhD graduation, I realized though that I'd be way happier and probably more impactful if I searched to industry and work in tech. So ideally at a company where machine learning is essential to their business, and I could learn a lot and also contribute. I had an Alexa Echo Dot at home, and I already had some growth idea opportunities for her. So, I thought, well, maybe I should join Amazon and actually start working on Amazon Alexa. So here I am.

Mary Killelea: That is amazing. I love that story. And I bet you pitched that in your interview.

Parmida Beigi: I did. I sure did.

Mary Killelea: So I guess AI, machine learning, data science is so complex. Another term that I've heard is natural language processing. What is that?

Parmida Beigi: Yeah, yeah, totally. Like all these terms, they're like really overloaded and we need to demystify, especially for those who are starting their career, and they just hear these buzzwords or even the public. Like, yeah, I really, I think it's a great question. So natural language processing or NLP for shirt is the ability of computers to understand natural language, which is the human language in spoken or written form. NLP is an interdisciplinary field of AI, linguistics, and computer science. And it's been around for more than 50 years. And maybe some examples would help. Like some examples of NLP would be language translation or writing articles or summarizing an essay or Q&A. And in fact, chat GPT, which I'm sure that most of your audience have probably played with is an NLP model. More specifically, it's a type of large language model, which is the state of the art for NLP.

Mary Killelea: Very, very interesting. Is there, moving and shifting to your job at Amazon? Can you give us kind of a synopsis of a day in the life?

Parmida Beigi: Of course. Yeah. As much as I can share, you know, I can share a secret matters, but yeah, totally. And I'd like to first mention that my experience would be definitely different from different orgs within Amazon. Like we have Alexa, AWS, or even different teams within the same org might be, you know, they day to day life might be different. But for our team, if we're not designing or releasing a new model, a typical day usually starts off with the customers. So as our most important leadership principle at Amazon also says, leaders start with the customers and move backwards. So, we study customer pain points and identify opportunities where we can improve customer experience. Then it would be the design phase where we plan how we're going to go about improving our CX or maybe we want to add a new feature. And all of that would definitely, we need to design and come up with a plan. And after that, after we've identified what the business problem is, and then we translated it to a science problem, now we can go ahead and solve it with data science techniques.

So we start up with data extraction, data exploration, visualization, data prep, finally modeling, lots and lots of unit tests, offline tests, A-B testing to confirm that the new model is definitely outperforming the old model based on the metrics that we're tracking. And then deploying, then post deployment, and looking for drifts and features, the model, then data, and of course, lots of meetings with stakeholders along the way. So yeah, maybe like, this is basically the end-to-end pipeline and how it would work, but maybe I spend maybe a week even just prepping the data.

Mary Killelea: So, that is so interesting. And I love that you walked us through the end-to-end because I think that's what so many people who are on the outside looking at building a career within don't quite understand all the touch points. So that was super helpful. What is it like for you personally to know that you're designing something that everyone uses, from your grandma to young children growing up right now?

Parmida Beigi: It feels incredibly rewarding and fulfilling to know that the work I do has an impact on so many people. Being able to work on something that millions of people use and interact with daily is a huge responsibility, but it's also an exciting challenge. I evaluate the performance of new models on my own Alexa device. And when I see the improved customer experience upon my model release, it's such an incredible feeling. And yeah, working on a product where I could directly impact the end customer is something that I've always wanted. And working alongside a group of talented individuals at Amazon, it just makes it such a unique experience.

Mary Killelea: Okay, so when you're developing a career strategy, it's hard to know, as I kind of talked about a few minutes ago, where you want to go if you don't know what all your options are. What advice do you have to help listeners understand their data-related roles and how to choose the right role for them?

Parmida Beigi: Absolutely, that's a fabulous question. And I could take the entire time of podcast to talk about that. But I try not to. So, in fact, a key difference between data science and other fields is that data science is not really well defined yet, unfortunately, which means that depending on the industry, or company or even orgs within the company, you may get a totally different expectation in terms of competencies. In computer science, for example, SD interview may consist of coding competency, system design and so on and so forth. For computer science, though, you may have a company that asks you, system design, or ask you to, you know, pass the SD one bar coding exam, or need you to be able to work with the cloud or have a PhD. And then there is another company who doesn't give you any coding interview, they don't require you to have a graduate degree, and they may just ask you for, you know, Tableau or Power BI visualization.

So yeah, so and there is a common trend that is that due to the hype around data science and AI, both applicants and employers are eager to market their roles as AI or data science positions, as this can give them a competitive edge. And we also have other data roles, it's not just data science, we have data engineering, data analysts, and those are, of course, better defined, it's not like data science, but sometimes they may still fall into the trap of the trend above, maybe it's a data analyst role, but people, you know, the employers just market it as data scientist. With that, it's crucial for applicants to do their due diligence, research the company to understand their commitment to data science and AI, as well as how essential data is to their business. So, it's more like, you should clarify whether they're more on the analytics side, like business intelligence side of things, or on the ML and deep learning side. So my advice is to not rely on the title of the job only. Definitely research the company, read the job description, be mindful of the various definitions that are used in different companies. Data science might actually mean just glorified data analytics, as I said, but machine learning, research, applied or full stack data scientist could be in fact, the true data scientists that we used to have like a decade ago. So use your judgment and the more job postings you examine, the more experience you'll get in classifying all these data roles into their proper category.

Mary Killelea: Great advice. So, if listeners don't currently follow you on Instagram, LinkedIn, and Twitter, they are really missing out because I am telling you, it's amazing information that you have in these snackable things. I've learned so much about AI and machine learning just through your posts. You use simple graphics and quick clips and you go by the handle big data queen, which I love. What got you into using and leveraging the social platforms to help women understand not only the differences, but a career path within those?

Parmida Beigi: Yeah, absolutely. And thank you so much for that. It's amazing to hear that I'm making an impact, a good one, I hope.

Mary Killelea: Definitely.

Parmida Beigi: And that in fact has been my goal since the beginning to you know, give back to help the community. It's been my goal in life basically since I was a kid. I remember it was a rainy evening in Vancouver, and I was going back home from work and that was when the idea of starting a public page and sharing my experience just sparked. And I wanted to give back, I wanted to share my anecdotes, help those who were just starting out or were confused in their tech journey. And yeah, so I started it. But it's definitely been a learning curve for me, as I'm a private person. But yeah, I'm really happy that I started it. I love the community, I get a lot of energy from their reactions, comments, DMs and the love they share. It's such a great experience.

Mary Killelea: Yeah, that's wonderful. And if you want to become a data scientist and are building out your resume, what are some of the soft and hard skills that they can accentuate? Because I know that's one of the things that you touch on in social media.

Parmida Beigi: Yeah, that's another great question. So, I can start with the hard skills. Again, data science is a broad and highly interdisciplinary field with overlapping soft fields, including math and statistics, machine learning, computer science and domain expertise. True data scientists need to have a diverse set of technical skills, including data analytics, programming and software engineering, deep learning like PyTorch, TensorFlow, and to end machine learning pipeline, even like MLOps and state of the art methods for their own niche. Like if it's NLP, they should be familiar with attention based models also for computer vision. They're the state of the art. And also model deployment and post deployment to monitor and look for drifts. So that's for true data scientists. But I also like to talk a little bit about the data scientist roles that we commonly see in the market. And that is data science generalist roles. So, it's different from true data scientist role. And the tech stack is definitely different as well.

So for data science generalist role, definitely the same data analytics stuff that still is the same like Python, NumPy, Pandas, libraries. SQL is definitely a must for anyone who's in a data related role. And then classical machine learning. So probably it's best not to focus too much on deep learning for a generalist role. Just classical ML techniques like even logistic regression or boosting techniques like GBMs, XGBoost. You'll be surprised how many industries have their production models trained on classical models still. And that's still so important. In fact, I recommend everyone, whoever I'm mentoring, to start with classical ML and really learn the foundations and not jump into deep learning. Although I know it's hard. It's like there's so much hype around it. So yeah, so that would be for the hard skills.

And for the soft skills, the first one would be problem solving. That's super important. Like if we wanted to explain data science in two words, that would be problem solving. To have this mindset of identifying problems, translating your business problem into a science one, diving deep into the issues and make data driven decisions to solve your problems. Next would be definitely communication. No matter what type of data scientist you are, you would need to frequently work with stakeholders and should be able to effectively communicate complex ideas and findings to both technical and non technical stakeholders. And the last one would be curiosity. Specifically as a data scientist, you need to develop this mindset of being just curious, just question, wonder why things are the way they are and ask clarifying questions. And yeah, that would be it.

Mary Killelea: I love the soft skills that you point out because I think so many people think for a technical role, you're only, you know, have a technical mind. But today to be effective and to, I think, work with various stakeholders across different orgs, you do have to have these important soft skills. So I'm so glad you mentioned those. What is the most asked question you get from women wanting to build a career in artificial intelligence?

Parmida Beigi: So the most asked question I get in general is how to get into data science that we talked about already. Now, the most asked question I get from women is actually about gender gap and negotiating salary and benefits effectively. So financial literacy is one of the key requirements. It's important for both men and women to have a good understanding of personal finance, budgeting and investing. And yeah, everyone should know their market value. They should do their research, their due diligence, and act smart in the entire process so that their package that they're given by a company is what they truly deserve.

Mary Killelea: What level of degrees are required?

Parmida Beigi: Yeah, that's a great question. So it really, again, we have two maybe, I mean, we have many types of data scientists, but let's just summarize it into two. One is data science generalists. The other one is full stack data scientists, which is the quote unquote true data scientist. So for two true data scientists, that is typically marketed as again, full stack research applied or machine learning scientists, you would need a graduate degree like master's or a PhD. A lot of the times the job description only mentions PhD actually. And that degree has to be in a technical field like, you know, computer science, engineering, math or statistics. And for data science generalists role, which is the most common type of data science, bachelor's even in a non-technical field typically is enough.

Mary Killelea: Help me understand. And I know you kind of touched on it a little bit before, but I want to go a little deeper. The difference between a data analyst and a data engineer and the tech stack skills for each.

Parmida Beigi: Great question. Yeah. While data analysts and data engineer are both data roles, their job responsibilities and skill sets are quite different. So data analysts work with data to extract insights and answer specific business questions. They use data visualization to identify patterns. They may use tools like Excel to run some aggregation and high level statistics. They may use SQL and they may also use Python, but like very light just for simple scripting.

Data engineers though are responsible for building, designing and maintaining the infrastructure that supports that data analytics. So, tasks like database management, data warehousing or data pipeline development. So, engineers need to have a strong understanding of computer science and software engineering. And they work closely with data scientists and analysts to ensure that what they need is available and is organized in a way that's good for them to just ingest. So they work closely together, but their responsibilities are totally different.

Mary Killelea: I know we're talking about all these various data related roles. Do non-tech companies have a need for these roles as well?

Parmida Beigi: Oh yeah, absolutely. Absolutely. A lot of data science and AI has grown into various industries like finance, retail, healthcare even, entertainment, marketing. And yeah, so they definitely have such roles. And actually there are really exciting ones like personalized medicine, which can help doctors to diagnose and treat patients more accurately. And then we have personalized advertising, which can help businesses target consumers more effectively. And that's one of the amazing things about a profession in AI or data science, that your domain expertise could give you a competitive edge when applying for a role. And specifically when you have another applicant who doesn't have that domain expertise, definitely super important. And yeah, so domain expertise, like business acumen, these two are essential components and easily accessible for anyone. You don't necessarily need to worry about the tech stack, but with only these two, being able to identify science problems from the business problem at hand, it's really important and crucial for this work.

Mary Killelea: Getting such good, good insights from you today. What kind of pay ranges, since I know this was one of the popular questions that you get asked, pay ranges could someone expect in data science roles? Let's just keep it broad with early in your career, mid and top in your field. And I know that's a hard question because there's so much that it depends on, but if there's a way to answer that in general terms, just to get the concept down. I mean, from marketing roles, it's pretty structured. You have this range, mid-career, almost across a bunch of different marketing positions. So, I don't know if it's like that with your area of the business.

Parmida Beigi: Yeah, yeah. It's such a great question and a difficult one. As you said, it depends on it, like it's thousands of factors. And I prefer not to actually give a range if that's okay.

Mary Killelea: Yeah, yeah.

PB: But instead I talk about all the variables here and I also have two really good websites that I'd like to suggest for people to find their market value. So when we're talking about pay range, we definitely need to pay attention to the industry. That's definitely important. The company, the location. So some certain states in the US, for example, pay better because of higher cost of living. California, where I'm at, is one of them, obviously. Then type of role, like what kind of data scientist is this role for? And it also varies significantly based on the candidate experience and the job level they are applying for. So after considering all these and many more factors into account, there we have a range for that particular level of the job. And then several factors like the candidate experience, as I said, but also if they have a competitive offer from another company, could decide where in that range that candidate fits.

So, with that, I'd like to suggest a couple of websites where you can search for all these, like location company. The first one is levels.fyi. And the other one is blind. So, these are two really good websites for this. You can estimate your market value, learn about different companies based on their location. And you can just add all the variables that you'd like. You might see some outliers there too. Some folks may just enjoy messing up with their stats on salary ranges and everything, but yeah, you'll get an idea.

Mary Killelea: That's awesome. And I'll include both of those in the show notes. So wonderful. Are you seeing more and more women getting into this field or what do you think still needs to happen to attract more women to the field?

Parmida Beigi: Yeah, I definitely see more women getting into AI and ML and data science compared to before, but there is still a lot we can and should do in my opinion. So gender gap, as I alluded to, still exists. Definitely stereotyping and unconscious bias is something that hiring managers and all of us need to improve on. We should address the gender bias in hiring, maybe by using blind hiring, maybe. That bias may also be there for promotion or performance evaluation. Fun fact. Well, not fun at all, actually. When I first joined the industry, I couldn't believe, and it was so disheartening to see that talented folks, unfortunately, especially women, being constantly talked over or having their ideas dismissed. And tech specifically is still a male dominated field. And there's a lot that has to be done, like better educating hiring managers and HR professionals for DEI, diversity and inclusion best practices, and creating an inclusive environment where you're welcoming everyone's opinion and giving equal opportunity to everyone.

Mary Killelea: Such great advice. In the news today, we've got some top tech execs and a group of artificial intelligence experts calling for a six-month pause in developing systems more powerful than OpenAI's newly launched the GPT-4. It's in an open letter citing potential risks to society. What is your point of view on, one, the need to take a pause and, two, the responsibilities that companies have and should be aware of?

Parmida Beigi: Yeah, absolutely. So, this letter and the requests come across as naive and unreasonable, honestly, but I definitely resonate with the concerns. They are valid. Like the fact that powerful AI systems should only be developed when we already have regulations in place for data and model usage. And when we have studied the model's ethical implications and have already developed the required guardrails to mitigate the potential malicious use cases, similar to how we gave ourselves an ample time to practice responsible software engineering.

In my opinion, we haven't taken the same approach with AI. And generative AI specifically, it has made so much positive impact across industries like healthcare, for example, education, and stopping the progress is not a viable approach. In my opinion, we need governments to step in to help. I mean, it's my opinion and also Andrew Nick's opinion. So EU, for example, introduced GDPR, and that was five years ago. It is a regulation for user data protection. And from what I gathered, there is also an AI act there, which is the equivalent of GDPR for AI, which is coming out this summer in Europe. And in the US, though, we are way behind. So, yeah, I think we need collaboration between experts in AI, policymakers, lawyers, and stakeholders to transform how we develop responsible AI. But the genie is already out of the bottle, as they say, and there is huge value in these AIs. And stopping is not a viable approach, in my opinion.

Mary Killelea: Yeah. What would you tell your 20-year-old self?

Parmida Beigi: I would tell her to pursue your passions. Don't be afraid to follow your heart and pursue things that truly inspire you. I'd remind her that challenges are opportunities for growth and learning, and embrace them with curiosity, openness, and don't be afraid to take risks, even if they feel scary and outside of your comfort zone. I would tell her, use your skills and talents to make a positive difference in the world. Find ways to give back to your peers, to your community, and use your skills, your compassion, your education, whatever that is, to make a meaningful impact. And finally, I'd tell her to remember that life is about the journey, not the destination.

Mary Killelea: That's beautiful. What does to be bolder mean to you?

Parmida Beigi: To be bolder, that means having the courage to stand up for your ideas, for yourself, to speak up, to take risks. It means making your voice heard when all the other voices around you are louder. It means to challenge biases and discrimination, break through barriers, and show the real you, the amazing person that you are to the world.

Mary Killelea: Fantastic. Just a couple more questions. What podcasts on AI do you recommend?

Parmida Beigi: Yeah, I like the AI podcast by NVIDIA. The way the stories are made and narrated are really nice. And also TwML AI podcast by Sam Cherrington and Practical AI by Chris and Daniel. They're current with what's going on the field and the conversations are just like really fun to listen to.

Mary Killelea: Great. I'll add those. What are some of the best certifications or training resources that you can share with the listeners?

Parmida Beigi: Of course, there are a lot of them actually. If I'm to pick a few, I like the machine learning specialization by Washington University. I love it. An amazing woman is actually teaching that. Oh, well, half of the course, I believe. I took the course like ages ago. But yeah, an amazing prof is teaching it. It's more on the technical side and you'll learn everything you need to know on conventional ML. And Rice University also has a good specialization with the same name. And I obviously have to mention the classical ML and deep learning specializations by Andrew Nick.

Mary Killelea: Fantastic. I could go on all day with you. You are just so refreshing to hear such an intelligent woman talking about AI, data science, machine learning and making it so approachable and understandable. Thank you for being here. Is there anything you'd like to say in closing to women listening?

Parmida Beigi: Yeah. So first off, it was my pleasure, Mary. And thank you for being a great host. And yeah, to women, I'm going to say we're in this together. I feel like the world is becoming a way better place than it used to be. Like even before when I was born, when I read articles, it's just so fascinating to see how far we've come. And so yeah, just keep going and believe in yourselves.

Mary Killelea: Thank you so much.

Parmida Beigi: Of course.

Mary Killelea: Thanks for listening to the episode today. It was really fun chatting with my guests. If you liked our show, please like it and share it with your friends. If you want to learn what we're up to, please go check out our website at 2bbolder.com. That's the number 2, little b, bolder.com.