Google, LinkedIn, Microsoft, ... and now EPFL PhD. Vinitra's Incredible Journey

This EPFL student interned at Google & Linkedin in high school. In uni, she interned at IBM Research, Project Jupyer & more. After 2 years at Microsoft, she left to start an EPFL PhD. Meet Vinitra.

Jan 29, 2024

On top of the impressive description, Vinitra has done much more. She scaled the Data 8 course up to 1400 students during her studies at UC Berkeley. She also graduated as the youngest masters student! And more recently - she published a paper at NeurIPS, and also worked on the world’s first Open LLM for medicine.

👋 Say hi to Vinitra Swamy (26), 4th Year EPFL PhD and a serial “excellence” achiever

If you’re not already subscribed, do it now so you don’t miss out on our upcoming speakers from Apple, Google and more!

Without further ado: This week I had the chance to chat with Vinitra, a super fascinating EPFL PhD. I tried my very best to make a summary of her countless achievements 😅

🏆 A track record of achievements

While she was at high school, Vinitra interned at Google HQ through the CAPE program, where she met influential figures (like the creators of Python and Java!) and she discovered her passion for Computer Science. She founded “Bridging the Digital Divide”, a project to teach digital literacy to those in housing shelters and retirement communities. When she ran out of CS options to do at high school, she took community college courses (which also transferred to her undergrad). She also joined the First Robotics Challenge, where she built and competed in national robotics competitions with an all-girl team (at a Nasa Research Facility 😮). She then interned at LinkedIn as a software engineer (before uni 🤯).
Vinitra did her undergrad at UC Berkeley, where she began TAing in her second year for the new Data Science course, receiving many awards and making significant contributions to project Jupyter and scaling auto-grading. She also interned at IBM Research by the end of her bachelors. Another notable achievement, she was able to complete her bachelors in just 2 years.
She continued at UC Berkeley for her masters, where she worked on enhancing the data science education. All this while being involved in many student orgs (like president of the Computer Science Honor Society) and also researching at the UC Berkeley RISELab. She held the achievement of being the youngest Masters graduate in CS from UC Berkeley!
After UC Berkeley, Vinitra went straight into the industry, working as an AI Engineer at Microsoft. Here, she worked on the Open Neural Network Exchange, to create a standard for neural networks so that engineers could translate between frameworks.
After 2 years at Microsoft, Vinitra decided to go for a PhD in Europe, with a focus on explainable AI and neural networks. She decided to join EPFL, and currently works with Tanja Käser (ML4ED Lab) and Martin Jaggi (MLO Lab). Recently, she published a paper at the NeurIPS conference, and she was also part of the team working on the very first Open 70B LLM for Medicine (two very cool achievements!). That’s not all - she’s also been working on other projects, being involved as President of the EPFL CS PhDs (EPIC) and also in Wingman Campus Fund where she led a start-up investment into Adaptyv Biosystems (a biotech start-up that went to Y Combinator).

Vinitra is very clearly an exceptional person and super passionate about education, teaching and computer science. It’s one of the reasons she popped up directly in my head when I was thinking of people to feature for this newsletter!

There’s so much to talk and learn about her different experiences, motivation and ambitions, here’s how our conversation went:

Motivation and Ambition

How do you think being a 2nd generation immigrant affected your achievements?

“I think my parents had a perspective that whatever you do, be the best in it, take advantage of all the opportunities you have. Whatever you're passionate about, do it 100%. I think that mentality is probably very similar to what your parents had too.

My parents were living in Michigan before I was born and moved to California, away from a lot of their friends and their support system. At that time my Dad took a new job at Apple and my Mom was pregnant. Knowing that they made that change to give us the best possible education is really meaningful. I feel it’s very important to put in the hard work and feel proud about our cultural heritage, especially seeing the support they give to me and my siblings.

It's probably also why you see, especially in the Bay Area, a lot of second gen immigrants are doing really, really well.”

What's the main factor that pushed you to pursue all these projects?

“Well, from the very beginning, I’ve always liked connecting with people. And one great way that I found of doing that, luckily early in my time at Berkeley, was teaching. This was through helping people without a computer science background understand how coding can be applied to anything they felt most excited about.

The first research lab I joined was the Berkeley Institute for Data Science. And exactly at that time, Berkeley was starting to form a data science major and department. They had just rolled out the first pilot class, my first semester at Berkeley. It started with just a 100 student pilot. And when I left, we had 1,400 students in that one class a semester. Teaching that class was a lot of my time at Berkeley and how I ended up down this path.”

You were involved with teaching but also very involved with Project Jupyter. Could you tell me more about this?

“Using Jupyter, we were trying to remove the command line cost of learning computer science. For instance, people from a different discipline can kind of get intimidated in an intro CS class with even just installing the packages, getting everything to work, getting a program to run for the first time, how an IDE looks... But computer science is about so much more than that.

Data science, I think, sits at the perfect intersection for this. A lot of my master's thesis and my work with RiseLab and with David Culler, who at that time was the Dean for Data Sciences, was spinning up JupyterHub for education, a distributed cloud based version of Jupyter with auto grading, authentication and interactive extensions, and managed compute environments on this gigantic scalable cluster. For students, this meant that they could do data science with any device that could connect to a browser.

At the time, Berkeley was the first university to scale this. A lot of things were on fire all the time because you had live student users that were submitting assignments at 2 a.m. and then the server would crash and you'd have to figure everything out.

It was very chaotic. No better place to try something like that. And also I think the scale was also what made me interested in startups and entrepreneurship. There's no where else where you go from 100 to 1400 student classroom and then in addition to that, we had a MOOC with another 75,000 students enrolled and we had five more courses that were also using the same cluster and all these connector clusters.

Everything was growing rapidly and now the department exists, which is amazing. But it was just the coolest thing to be involved with and it was really only possible in those early years.”

What did you find most interesting? Was it the teaching part of the course or working on scaling JupyterHub?

“I was always kind of involved in both sides. I think the Jupyter ecosystem is a nice technical challenge, but what makes my heart happy is really getting people their first introduction to computer science in a way that makes sense to them and makes them feel excited and inspired about what can be done and what they're passionate about.”

You’ve done so many projects over the years, many in parallel. How do you juggle all of these?

“Whenever a certain project gets hard or you're feeling unmotivated to do it, switching is always nice. So I think it's always good to keep two projects on at any point in time. You can keep switching them out.

But yeah, I felt very passionately about teaching. And I also really liked my team and my job. Whenever I, for example, was getting tired of coding a bit and wanted to context switch into something more conceptual, I would think about how to frame an idea in one of my lectures. And then when I was tired of this, I really wanted to go into something practical. I would switch back to coding. So maybe I need both. Maybe that's the answer.”

Working in the industry

You joined Microsoft as an AI engineer right after your Masters. How was your experience?

“I loved it, honestly. It was really, really, really nice. I was on the AI Frameworks team within the Microsoft AI + Research division at the time, which has now since split off into Microsoft Cloud and AI. It was a really unique team for a couple of reasons.

One, the team was working on open source AI. That was really important to me because I think it really means a lot to be able to contribute to the community globally, even if you are working in big tech and not just in the form of products.

The second thing was that the team was quite special, it had a mix of both researchers and engineers working on the product side. So it was really a unique balance. And I think that this is exactly what you need to be on the cutting edge of AI, both how to implement the features appropriately, but also working on the next best thing or the state of the art.

And I think the culture at Microsoft is really really nice; the people, the company, they really take care of you. And Seattle is also a beautiful place, although rainy all the time. But the summers are particularly nice because everyone is so excited that it's sunny. All of those factors came together to have a really nice experience.”

As a student, we always have this big decision to make between industry and academia. You actually experienced both first-hand. I would love to hear how you navigated this.

“Yeah, well, actually after my master's, I never thought I would do a PhD. I was like, ‘Industry, take me! I want to work on hard problems at scale that impact tons of people!'

Industry has lots of exciting new things to learn, ways to contribute. I loved being part of a large organization and meeting lots of interesting, intelligent people, especially while presenting our work at conferences. The scale of industry resources, as well as ability to move quickly are unmatched.

However, about a year into Microsoft, I started feeling a bit restless. The work had stabilized a bit after a few large releases. The challenge became a bit easier. I realized I wanted to work on moonshot ideas and I wanted to teach again, both of which are much more possible in academia. I loved the team and I loved working on open source, but I felt it was time for the next adventure.”

Ok, so then you decided to pursue a PhD. How did you end up picking EPFL?

I was telling you about that golden handcuff feeling, right? I could see myself in five years hopefully making a lot of impact at Microsoft, but not really working on what I felt was the most pressing thing to work on.

I think later in life, when you maybe have a family and when you've settled down a bit, these kinds of jobs make a lot more sense with a stable career and impact and scope, roles. But at that point I was like quite young, I guess 21 or 22 (Editor note: 🤯).

I was very ready to go all in on something that was like crazy and unstable and very passionate working all-nighters with people that felt very strongly about it. And exploring the world and traveling and all of this.

So I only applied to European PhDs. I got a number of offers, which were really nice, but I think that was the time the pandemic started. All my open houses got canceled. I wanted to visit schools and make my decision based on how I felt about the people, etc. Instead, we had Zoom calls, and I felt strongly that the PhDs at EPFL seemed the most happy and balanced.

Also, the professors here seemed the most excited about what I was proposing, AI for education, but also specifically from the technical standpoint, my research interest in explainable AI and how we can understand neural network decisions.

There were not a lot of people working on it in Europe at the time, almost none. So you kind of had to convince people that, ‘this was interesting and we should work on this together, even though your lab doesn't do this right now’ which is quite a call to make.

And both Tanja Käser and Martin Jaggi here at EPFL were super enthusiastic about it. They were willing to say, ‘we will learn together on this’. And I thought that was just a really, really nice perspective, so that’s what led me to EPFL.”

LLM and NLP Research

You’ve published to NeurIPS recently, congrats! You’re also part of the team that launched Meditron-70B, an open LLM for medicine. What are your thoughts on the future of LLMs?

“Thanks! That's a big question, yeah. I think language models have done something very, very important, which has made AI accessible to the everyday person to see what it's capable of. And I know some people are scared of it and some people are very excited by it.

I think it has shown people very clearly what kind of tasks can be automated and what kind of tasks we really need humans for and how humans and AI can work together to collaborate, to create some amazing things.

I don't think AI is at the state (and maybe will ever reach the state) where it can reason more creatively than humans, without some huge technological advances.

What I do think is really awesome is that a lot of unnecessary mental labor like repetitive filing of things, etc, those kinds of tasks can really be taken care of with a higher accuracy and with not very much effort by AI, which leaves space for humanity to pursue so many more creative and higher level thought processes.

I think, we as a world will advance in terms of our intellectual curiosities, capabilities and contributions because of this.

From the education perspective, I think our schooling curricula needs to change to account for this in the way we test and the way we teach. There's a lot of work to be done even with the current advances, and I don't think it's been fully exploited yet. The next four or five years will show us exactly how many different ways LLMs can really help when it's integrated into our daily lives.”

Future Plans

What are your plans for later?

“I'm still figuring it out. I think end game I would like to teach data science. Right now, probably right after my PhD, I think the education space is primed for a revamp.

I’m hoping to found a startup in ML for education. I really think that there's a lot of good to be done here. I don't want to be yet another education platform, I want to tailor student learning experiences on existing platforms. I want to make exams far more intuitive. I want students to feel like they can connect with the material. And I believe all of this is possible with AI. The future is exciting!”

Advice for the reader

Is there any advice you'd give to the bachelor and masters students reading this?

“Find something you're passionate about and go all in on it. It doesn't mean that's what you have to do for the rest of your life. But I think that getting that depth is really nice.

The other part of it is to not let anyone tell you that you can't do something. I think a lot of people hear this abstractly, but if someone doesn't say a direct no, then it means that there is a way that it's possible. I think this ‘ask for forgiveness instead of permission’ phrase is really important when it comes to trying crazy things like graduating early or applying for a position you feel underqualified for.

Just try it. Don't be worried about what people think. And good things can happen to passionate people!”

Closing notes

Well, that concludes the super interesting conversation with Vinitra. This is just the very first edition of our student spotlights, and so we’d love to hear your feedback in the comments down below, I look forward to reading them :) Also vote in the poll!

Stay tuned!

Arnie

P.S. And for the many of you currently in your exams, best of luck 🤞!

Ambitious & Driven