Practicum Spotlight
A candid conversation with three current MSDS students about their internship experiences at Hims & Hers and Twist Bioscience.
Background and Introduction
Karishma: Before coming to USF, I was a senior software engineer at John Deere in India, and there I worked on a data science project based on computer vision and machine learning. That’s really where I developed this interest in data science, so I decided to pursue my career in this field further. I came to USF because it’s famous for its coursework, faculty, and practicum component, so I decided that this was the best place for me to go. I started my practicum in October at Hims & Hers, working alongside Jason.
Jason: I didn’t really come into this program with a lot of work experience; I came here straight out of college. I did my undergrad at UC Santa Barbara where I was a stats major, and like Karishma said, I’m also currently interning at Hims & Hers. Being a stats major, I always had an interest in data science, and with San Francisco being in Silicon Valley, USF felt like the perfect place for a new grad like me to be starting out. With USF’s location and practicum component, it was a no-brainer for me.
Vishwas: Prior to the MSDS program, I was working in India at a company called ACG World, where I was working with a lot of IoT–industrial Internet of Things. These were IoT sensors that would be put into machines that the company would sell to customers, and we were building a platform for our customers and developing machine learning and data science solutions. Before that, I was working as a data analyst at Reliance Industries, so I would say that’s where my interest in data science really began. I was most drawn to USF’s practicum program, since I found the opportunity to gain work experience while being in grad school to be extremely valuable. Currently, I’m interning at Twist Bioscience, where we work on a lot of DNA sequencing data.
Projects
Jason: Karishma and I have worked on a lot of stuff at Hims & Hers. We started out the year making a TV advertisement baseline project, and that directly contributed to helping the marketing department determine if TV advertisements were effectively generating additional traffic. Right now, we’re working on another high impact project where we’d like to predict customer lifetime value for consumers. Customer lifetime value is essentially how much revenue each customer will generate by the end of the year. So, whenever we have a new customer or a new subscriber, we’d like to predict how much revenue that customer will bring to the company.
Evie: And just to clarify, could you explain what Hims & Hers sells as a product to its customers or subscribers?
Karishma: So basically a customer will come to the website, for example, if they have a health issue, and then they will get a prescription from the doctor, and that doctor is also affiliated with Hims & Hers. Once they have their prescription, the customer will subscribe to having those medications sent to them through Hims & Hers.
Jason: Just to add to that, the mission of Hims & Hers really focuses on is destigmatizing these different types of conditions that people face. You’re able to talk to a doctor about it in the privacy of your own home, and you don’t have to go into an office, potentially feel judgment from others, and can still get a prescription and your medication in a seamless way.
Evie: That’s awesome, thanks so much for sharing! Vishwas, what’s your practicum project like?
Vishwas: In my practicum, I work on a lot of sequencing data. One of my main projects that I’m working on is antibody discovery. We experiment with getting a lot of sequencing data, and then the problem we want to solve is finding out the sequences that actually bind to the target as those would be the antibodies, and we do a lot of this work with deep learning models. I’ve used models like CNN, ResNet, and similar models for classification, specifically binary classification. Twist primarily works as a B2B (business-to-business), so we support other research organizations as well, and our customers come to Twist with a request and Twist will manufacture that DNA for the customer. So that’s one branch of Twist, and then another branch is more research-oriented and works on the antibody discovery project that I just mentioned.
Most Useful Class
Karishma: For our first project, we used the Facebook Prophet model a lot to handle time series problems. As we were going through our time series class, this is when we were using a lot of those concepts as well, so it was really useful that it aligned with our project so well. Later, for our second project, we made our Phase 1 model based on tree-based models like XGBoost and Random Forest, so it was great that we could learn those in class and then apply to them to our projects really easily.
Jason: I agree. Overall the machine learning courses were super useful, especially Yannet’s machine learning lab.
Vishwas: It’s difficult to pick just one subject, because everything starts to line up as the courses go along. Linear regression in the fall was such a fundamental thing for us to learn, so that was really helpful. Of course, machine learning was really helpful, like knowing how to generate features, do validation and cross validation; and then in advanced machine learning, building an understanding of neural networks, PyTorch, and creating datasets for deep learning models. My practicum project really lined up with all of these courses, so they’ve all been really helpful. Even now, in this module with product analytics, it’s been useful to learn about deploying models, so I can apply that to deploying the models I’ve been working on at Twist.
Tools and Technologies
Karishma: Apart from the Facebook Prophet model and other tools that we’ve used in the MSDS program, we’ve also used DWT which is different, and of course, we use a lot of Python and Python packages to write our code. In Phase 2 of our project, we’ve been using RNNs and feed-forward neural networks.
Jason: Pretty much all of our code in done in Python; we live in Pandas and we live in Scikit-learn.
Vishwas: I’ve been working with Databricks and using a lot of PySpark–basically a lot of the tools we learn in the distributed computing and data systems courses.
Most Interesting Thing Learned
Jason: I was caught a little off guard by how unstructured data science feels, even in a professional setting, since it’s such a new field. Even with so many experts, sometimes we aren’t 100% sure what we should be doing, but it’s nice knowing that a lot of people feel this way and that’s why we work together to move the field of data science forward.
Vishwas: I learned that a lot of data science is about experimenting–sometimes you hit the target, and sometimes you don’t, and you have to figure out what works best and what doesn’t work at all. I also learned the importance of cleaning and preparing data — so much of the work that is done is pretty much doing that, and then much less time goes into actually building the models themselves. I think for any incoming data scientist or any incoming student of data science, they should definitely expect that out of data science as a profession. Lastly, sometimes it can feel scary to go into uncharted territory with a project in data science — that feeling was something else I learned through my practicum experience.
Karishma: I totally agree, data science can feel very uncertain at times. It’s not like you just do some programming and get your answer. It’s always relative, whether you want to improve a metric or incorporate a new type of data. Sometimes, you have to spend a lot of time on the data to get results, and you have to be ready to learn any type of model or tool.
Data Science Interests
Karishma: I really like the modeling and machine learning part the most. It’s interesting to work on improving models, getting results from different models, creating ensembles, or just learning about new models in general. I also like coding, so mixed with that, that’s where most of my interests are.
Jason: I feel the same as Karishma. I’m pretty passionate about the modeling side of data science, but for me, a specific industry I’d really like to go into is social media and tech. I’d love to analyze user data and user behavior and see what kind of impact I could make there.
Vishwas: I’d say working on data pipelines and building data pipelines, as well as the modeling part as Karishma and Jason mentioned. When you have to work on large datasets, thinking about parallelism and distributed computing and optimizing the data pipeline is so important. In the future, I could also see myself working in the biotech field, similar to what Twist is doing now.
Free Time
Karishma: I enjoy going to concerts, and I recently saw BTS. Besides going to the actual concerts, I love watching their videos and past performances. I also dance, so I love doing that too!
Jason: In my free time, I like to go to the gym, play basketball, and just hang out with my friends outside of school. I actually live with three other MSDS students in my house, and it’s really cool being able to ask my housemates for help on homework or just to hang out whenever we’re all free.
Vishwas: I love San Francisco — it’s such a great place to live and there are so many places to see, so I love going around the city during my free time. Whenever I have free time, I also try to go for a run around my neighborhood. Currently, I’m also living with five other MSDS students, so I can relate to what Jason was saying!
