Breaking Into AI: Sahar Nasiri on Acing the Data Science Job Interview

Sahar Nasiri, data scientist at Delta Airlines, shares her career journey.

Data scientist Sahar Nasiri originally went to college to study industrial engineering. After taking Andrew Ng’s Machine Learning course on a professor’s recommendation, however, she knew she wanted her future to be in AI. Now she uses AI to help Delta Airlines keep its planes in top operating condition. She spoke with us about her early interview struggles, how she landed her first job, and the value of truly understanding statistics.

Name: Sahar Nasiri

Title: Data Scientist, Delta Airlines

Location: Atlanta, Georgia

Education: Master’s of Science, Data Science, University of Rochester; Bachelor’s of Science, Industrial Engineering, Azad University

Hobbies: Theater, especially modern plays. 

Can you tell me about your current role? When did you start, what is your title, and what are your primary responsibilities?

I recently moved into a new department at Delta called Engineering Analytics, where I use NLP to analyze text logs generated by maintenance staff, flight attendants, and pilots.  For example, a flight attendant may write a comment in a log page that she smells something bad. Based on how she described it our program is able to flag her comment as a severe issue or non-severe issue. This optimizes the maintenance schedule and saves Delta a lot of money. Reliability specialists used to do this job by hand, and it was very time-consuming.

What does your average day look like? 

I spend a lot of time reading articles about transformers and sequence modeling, all pertaining to the specific problem I am working on. 

The other part of my job is leading a capstone project with a university that we are collaborating with. I try to get my hands dirty and review their code as closely as possible so I can help them improve. The students get real life experience, and they help us explore new solutions for our projects. They share a lot of good ideas. I will probably lead another project with another university next semester.

How did you get interested in machine learning and data science?

I originally went to school to pursue industrial engineering. However, my university had one of the best robotics teams in Iran. I wanted to join, so during my first summer of undergrad I took a robotics course. During one class, our professor showed us a presentation on Andrew Ng’s Machine Learning course from Stanford. I eventually took the course and thought, wow, this is what I want to do with my future.

What is the biggest thing you have learned about data science as a profession since entering the job market? 

I used to underestimate the power of statistics, and I lost a lot of job opportunities because of this lack of statistics knowledge. That made me realize that I needed to catch up.

When did you realize that you were lacking understanding in statistics and mathematics? 

I remember the exact moment. I was interviewing with Paypal and during the final stages they asked me what kind of graph I would use to analyze the relationship between two continuous variables. The answer was a scatter plot, so that it would show certain patterns, but I didn’t know that at the time. It was very basic stuff, but I wasn’t yet experienced with statistical experience in exploratory data analysis. In grad school, they did teach us how to do exploratory data analysis to plot different types of graphs, but they didn’t emphasize understanding when to know which type of graph to use.

Data scientist Sahar Nasiri at the Google headquarters.

Once you identified these shortcomings how did you go about catching up? 

I started by taking two statistics courses on Coursera, one from Duke University and the other from University of California, Santa Cruz. Both were really valuable. For instance, I always knew what p-value was in a theoretical way, but the courses helped me to really understand how it actually worked. 

Around that same time a friend from college moved from Iran to Canada, so she and I were in the same time zone. We made an agreement to start studying different algorithms together. We would pick a different algorithm each week and learn every detail about it using papers, data science websites, and videos from DataCamp, Kahn Academy, and a YouTube channel called StatQuest.

Tell me more about your study group. How did you schedule meeting times and coordinate assignments?   

At first we agreed to study the same algorithm independently throughout the week, and then call each other on Sundays to talk about what we had found. We soon determined that it was better if we actually studied at the same time. We would try to find an hour each day and be online together. During that time we talked about the paper while making notes. Once we finished studying each algorithm, we would go through all our notes and create a document to refer to when preparing for job interviews. We stopped temporarily because she is finishing her master’s and I am busy with my new position, but we plan to start back up again because we both learned a lot from it.

From your experiences, which algorithms would you recommend people focus on?

I don’t think companies expect you to know every single detail about every algorithm, but they do want you to know the algorithms that you put on your resume. You should also know some of the basic ones, such as logistic regression, decision trees, and gradient descent

In interviews, you want to be able to answer questions that ask you to choose one algorithm over another for a certain type of analysis, and explain why. For example, one of the most popular questions in my experience is: Why would you use a decision tree and not a support vector machine for a classification problem? For my last interview, I presented a project that compared several different classification algorithms together, along with their runtime and accuracy. I was able to explain why I chose the one I did, and which hyperparameters I tuned in order to get the best outcome. 

How did you finally land your first job?

My first job was also with Delta, but in a different data science role. For that role they weren’t really interested in whether I knew the mathematics behind everything. They only really cared whether I could program in Python, and I have always been good with Python and SQL. 

The first thing I did after getting hired was study all the things that I failed on during other interviews. I started with decision trees, then K-means clustering, and all the other core machine learning algorithms.

What sort of questions did they ask you during the interview? 

They gave me an unbalanced classification problem and asked me to explain how I would analyze it. This same question came up during a previous job interview, except pertaining to credit cards. I did not do well in the earlier interview because I wasn’t able to explain why my accuracy was low. With Delta, I was able to confidently explain all my analysis because of the studying I had done.

What was your day-to-day work like at that first data science job? 

My role was as a data scientist working in price optimization. Every Delta flight has many different possible prices for each origin and destination, so it is very difficult to find the one that makes the most sense to advertise. At the time I was hired they were hiring data scientists and Python programmers to shift the whole process over to machine learning for better optimization. 

I also worked on seasonality. During low season tickets are cheaper, and during high season they are more expensive. Previously, the season for each origin-destination was defined manually by analysts, which is a ton of work and really inefficient. We developed a process that goes through every origin and destination and finds the seasons for them and optimal prices in under two hours.

Do you have any advice for people aspiring to become data scientists? 

I always tell people to never stop learning statistics. If you want to be a data scientist you will always need to know statistics. 

I think anybody can learn to use machine learning packages in Python or R. What makes a data scientist different from others is their ability to explain the models they choose and how to tune the hyperparameters. I think studying the mathematics behind algorithms and the logic that they are following is necessary for us to thrive.

For many people, life feels too busy to schedule daily, or even weekly, study time. Do you have advice for how to fit learning into your day? 

Growing up, my dad was obsessed with time management. When I was 12 he took me to a seminar on scheduling. The speaker said something I’ve always tried to stick to: Don’t keep planning to do something if you aren’t going to stick to it; You are only going to exhaust yourself and lose your motivation to do anything else. If I’m trying to force myself to study but never able to make the time, I recognize that I’m not mentally ready to do that right now. I simply take a break. Once some time passes I will try to pick it up again.

As an example, during one of my job searches I was having trouble sticking with my study schedule. I first tried cutting it back, from an hour a day to 30 minutes. I still couldn’t focus, however, due to the stress of the job search. I decided to give it up for the time being and promised to pick it back up once I got a new job, which I did. Give yourself a break, but know that you still have to try and return to it.

Data scientist Sahar Nasiri (far right) with colleagues.

You immigrated from Iran to the US. Do you have advice you can offer to other professionals who emigrate for work?   

Getting a work visa can be very hard, especially if you are coming to the U.S. It is easier to get a student visa. However, as a student you aren’t allowed to work off-campus. This gives you a narrow window to find a real world job before you graduate. I was having trouble getting any interviews by just applying and sending in my resume because hiring departments just adds yours to a pile. 

So, I went on LinkedIn and reached out to everyone I could find working at companies I found interesting. I would say something like, “I saw a position is open at your company. If I gave you my resume can you act as my reference?” I would email about 200 people a day, and roughly 40 percent would respond. 

Our industry has a lot of immigrants, and many of them sympathized with me and were very helpful. After having an interview, no matter how the interview went, I would say thank you and try to stay in touch. I also try to do the same for others. I give referrals when people ask me on LinkedIn. I will also review resumes for anyone who asks.

How welcoming has the AI industry been to you as a woman? 

I think it’s a great time to be a woman in tech right now. Everyone is interested in increasing diversity so if you are a woman people are more willing to give you a chance. We have our own conferences, too, like the Grace Hopper Celebration every year in Orlando. This is a great resource to network with 25,000 other women in our industry, working in companies from all sizes.

Do you have any other advice for people hoping to start careers in AI? 

I always tell people that the best starting point is to learn to program in Python. If you want to be a data scientist, you must never stop learning statistics. 

What is your next career goal, and how are you working to achieve it?

I’m trying to become an expert in transformers, because they are taking over the NLP world. I read and take courses for 45 minutes each day. To start, I am studying Francois Chollet’s amazing book Deep Learning With Python. I have also finished 4 out of 5 of the courses in the Deep Learning Specialization and have also been taking the NLP Specialization

I read Jay Alammar’s articles on NLP, which are all amazing, as well as any relevant articles on towardsdatascience.com and medium.com. The rest of my time goes to implementing things I have learned while doing my projects.

Connect with Sahar Nasiri on LinkedIn!

Want to strengthen your knowledge and ace every interview?