A conversation with Sameera Poduri, our Head of Data Science
Sameera Poduri joined Amino in January 2017 to lead our data science team. She has a PhD in computer science, and her areas of specialization include machine learning, A.I., and robotics.
What inspired you to join Amino?
I first discovered Amino as a user. When my family moved to San Francisco, I was struck by the contrast between the process of finding a new doctor and the process of finding a new place to live. For the latter, I could easily look at videos of apartments online and get a dollar value upfront, whereas finding a doctor was a difficult, confusing process. I didn’t have a doctor secured by the time we moved and, sure enough, my daughter fell sick our first week in San Francisco. That’s when one of my friends showed me Amino. I remember that feeling of relief and the data scientist in me thinking “this makes so much sense.” So I used Amino to find a doctor and book an appointment.
A year later, I heard from Amino about the head of data science role, and I had to check it out. I was impressed with the vision and thoughtful approach of the leadership, and also with the diverse group of talented people they had brought together. That’s how I joined!
Can you give a high-level overview of your role as Head of Data Science?
My role is a combination of strategic and technical leadership, and team management. I work closely with the product, sales and, marketing leaders to align the data science team and execute on the company’s product roadmap. I spend a lot of time thinking about data science processes, cross-team collaboration, and efficiency. I provide technical guidance on the challenging problems the data science team is working on, including research on new and innovative data products. I also really enjoy mentoring my team members and helping them grow in their careers.
What is the data science team responsible for at Amino?
The data science team at Amino builds the algorithms behind the product’s core features, such as doctor and facility recommendations, cost estimates for 100+ procedures, quality measures, and experience summaries for doctors such as the C-section decision factor. Our process typically involves data exploration and research to identify interesting new product features, developing algorithms to enable those features, analyzing and validating the output, and finally, writing production code and monitoring the feature after deployment. We work closely with the product managers and the engineering teams.
What is an area that data science (as an industry) could improve?
Valuing simplicity. In the data science community, there is a lot of hype around AI and techniques such as deep learning. While I am personally very excited about these breakthroughs, I also see that because of the hype, there is a tendency in the field to apply complicated black box algorithms to problems without sufficient thought to the structure of the problem and data at hand, and what the product needs are.
Sometimes it is more important to focus on understanding the data and building models that are robust, explainable and easy to deploy and maintain—I think this is especially true for startups. My recommendation would be to start with simple models, explore more complex ones, and ask if the the performance gain is worth the additional complexity and deployment cost.
What is your favorite aspect of the job?
I think there are two axes when it comes to career fulfillment: being able to do what you’re good at and being able to do what you believe in. Personally, what I’m really good at is mathematical modeling and algorithm design. I enjoy connecting the dots between different algorithms that have come up in various contexts and being able to create a new algorithm for problems. I enjoy being able to collaborate on multiple projects with very bright data scientists at Amino.
What I really believe in is building useful products that people actually use. I love that I get that opportunity at Amino. We have a significant problem around healthcare transparency, we have access to really valuable data, and we are empowered to create a solution.
What is the most challenging aspect of the job?
Data science for healthcare is a huge opportunity, but also very challenging. We’re designing algorithms for novel data products and there’s no guidebook of algorithms—unlike many other verticals. For example, we’re trying to predict costs for hundreds of procedures, across thousands of facilities and insurance plans across the country. The data is messy with sparse labeled data sets and, yet, the products can really affect people’s healthcare choices, so we have to be very careful about algorithm performance. In addition to being accurate and scalable, the models also have to be explainable so our users can understand what we show in our product.
Another challenge is communicating our work across the company and with our customers. This is really important for data scientists—we’re the closest to the data, algorithms, and limitations, so it’s important to have open lines of communication where we can clearly explain what we’re doing and also receive feedback on the work we do.
What is your vision for the future of data science at Amino?
Data Science at Amino has a huge potential for impact—we have a very rich healthcare data set and are just beginning to unlock its value. We will continue building new and innovative data products that further Amino’s vision of connecting people to the best possible healthcare. As we do this, we are getting better at designing algorithms and analyses that balance accuracy, explainability and efficiency.
We are also getting very good at the data science process of going from exploration and research to prototype and then product deployment and continuous iteration on it. For data scientists looking to make a big and meaningful impact, healthcare is one of the best fields to work in, and I think Amino will play a very important role.
It’s a well-known fact that STEM fields lack diversity. How can leaders such as yourself help combat this problem?
Having a diverse group is necessary in data science in order to design algorithms that serve all people. The STEM industries are clearly facing barriers to achieving this diversity, given that the percentage of women in computer science is less than 20%, which is a serious problem. We should also be mindful of the fact that diversity isn’t only about gender, but also about factors like race, age, and even professional background.
I’m really proud that at Amino, our data science, data engineering, and product teams are diverse. I can see how having different perspectives shifts the dialogue—all the important points get raised, we’re more open to ideas and feedback that are different from our way of thinking, we learn more, and the quality of our work improves. This is especially important in healthcare, an industry in which it’s crucial to avoid biases in the algorithms that we create and the doctors we recommend.
Diversity is also really important to me on a personal level, and I work hard to ensure our team is thoughtful in the way we build our products, approach problems, and treat each other. There are lots of things other data science leaders can continue to do to move the dialogue around diversity in a positive direction. I personally encourage taking on mentorship roles, engaging in one-on-one conversations, and being visible at key events. We’ve made progress, but there’s still so much more to be done.