Analytics is like an onion: My internship experience at DSAID
Hello! I am Livana Ho, a fourth-year student majoring in Smart City Management and Technology from SMU. During my summer break in 2022, I interned with the Quantitative Strategy team in the Data Science and AI Division at GovTech. I was attached to a forward deployed team that is currently deployed to EnterpriseSG (refer to Joy’s article on what FDT is and what they do). Here is my honest take on what my internship journey was like:
How did I end up here?
As cliched as it sounds, the tagline: “Tech for Public Good” was what drove me to apply for an internship with GovTech. Despite not exactly knowing what career I wanted to do, I knew I wanted to devote my time to something that will benefit and impact people. An internship with GovTech felt like the best place to get first-hand experience of working at and with public agencies for the people.
My time here was spent on an analytics project that would help the agency improve its operations by providing users with data that empowers their decisions. Before starting the project, my teammate sent me a very insightful article that outlined the difference between data-driven decisions and decision-driven data analytics (cite). Indeed, it was not about finding the uses for the data on hand but being able to accurately translate business needs to the data we have on hand. I later experienced this struggle to change my thinking from the former to the latter.
What did I do and learn? What is this onion…?
When the analytics task was first assigned to me, it sounded straightforward: “Help the agency forecast X so that they can take more proactive interventions”. Easy peasy right? Just code out a forecasting model and we are done. Nope, that was not it. I soon found out that the analytics task had many underlying levels to it — like an onion.
That was only the first layer. The forecasted number was not enough as it only tells us what but not exactly how. We must push on to peel the next layer of the onion even if it makes us cry. The next layer is about further breaking down the forecasted values for the end users to understand how and what they could implement on the ground. Then the next would be to showcase the variables that cause this behaviour, and the next and the next and the next…
Else, you could also look at this onion from another point of view: to answer the high-level question of “how to take more proactive interventions on X”, the first layer would be to answer what the current behaviours are. Then the second layer would be to extract the factor causing those behaviours, how likely these factors are, what can we do to respond to these factors, and so on and so forth.
Unlike the project requirements we are used to for school projects, analytics in the real world is more than getting the best accuracy. It is the art of finding a good balance between results and translating it into business actionable. I am thankful that I had my team for being my bowl of water to guide me through the translation of analytics to business actions so that the peeling of the onion becomes less painful.
The technical side of the onion
My analytics project dealt mainly with time series data, which was, unfortunately, my first time doing so. As such, I spent some time learning and experimenting to understand the difference between typical machine learning models and time series models. In contrast to typical machine learning methods, which assume that the data points are independent, time series data are sequential and dependent on one another.
I learned about terms like trend, seasonality, cyclic, stationary and how it affects the type of time series model I can work with. Some of the time series forecasting algorithms I tried are Exponential Smoothing, ARIMA, Facebook Prophet and Long Short-Term Memory.
The feasibility of each model depends on whether time series data is univariate, has trend or seasonality and size of dataset. For example, simple exponential smoothing works well for data that do not have trend or seasonality, double exponential works for data with trend but without seasonality and final triple exponential works for data with both.
Apart from the forecasting, I was also tasked to create a mockup to allow end-users to interact with the models. I utilised an open-source app framework, Streamlit, which enabled me to quickly build and deploy the mockup. It was certainly fun to be able to see the models go from code to an interactive mockup that users can play around with.
What was it like interning here?
Since this was my first tech internship, I was not sure what to expect. Contrary to my expectations of doing trivial and mindless tasks, the task assigned to me had tangible impacts on the agency, which made me appreciate the struggles I faced.
In addition, I am very thankful that I was treated like an equal member of the FDT and given respect despite my lack of experience. My team had been very accommodating, from answering seemingly silly questions and giving constructive feedback to guiding me back onto the right direction for the project.
QS Culture
Fortunately, I was also able to experience the wonderful QS culture as we slowly ease back into hybrid working arrangement. There was bi-weekly sharing conducted by different QS members and it was surprising to see and learn about very different topics each time. This goes to show that the QS department is truly diverse, and people were happy to learn from one another. Apart from the ‘nerdy’ stuff, everyone takes time to play and enjoy themselves by going on evening runs, bouldering and even wine nights together!
Overall, this internship has been unexpectedly wonderful and insightful. I gained not only the experience of working on a real-world analytics project, but also what I did not know about myself that I would like to improve on. A final thanks to my team (Jovi, Jing Song), my RO (Shane) and the rest of QS and EnterpriseSG for having me!
If you are still on the fence about joining GovTech or QS, here’s your sign to do it!
Part of the Data Science and AI Division at GovTech, the Quantitative Strategy team works closely with public sector agencies on high-impact data science projects to enhance policymaking, operational work, and service delivery, all to support Singapore’s vision of building a Smart Nation.
We take in around 3–6 interns every cycle (Spring, Summer, Fall). In your internship, you’ll be attached directly to one of our project teams to work with them on an ongoing data science project. You’ll gain hands-on experience in delivering data-driven insights and in developing machine learning models for complex problems. Beyond sharpening your technical skills, you’ll also learn how to effectively scope data science projects and communicate with non-technical audiences.
Apply for our internship programme here today!
