Internship Experience in DSAID during COVID
This story was originally published by Fan Ting Wei on DSAID’s Medium blog.
I am Ting Wei, a third-year student majoring in Industrial Engineering in National University of Singapore.
The ongoing COVID-19 pandemic has placed great focus on the issue of protecting jobs and livelihood of Singaporeans. I am glad I was able to contribute to the government’s efforts to alleviate job losses through my internship with the JumpStart team in GovTech’s Data Science and Artificial Intelligence Division (DSAID), and I hope this blogpost about my twelve-week internship experience will encourage you to apply to GovTech to do meaningful work and develop tech for the public good.
Learning about Recommender Systems
Have you ever wondered how job portals recommend jobs to you? The JumpStart team in DSAID works on recommender systems for MyCareersFuture, a job portal for Singapore Citizens and PRs developed by Workforce Singapore (WSG), in collaboration with GovTech.
The team built and maintains recommender models that provides job recommendations for search results and job description pages. To recommend jobs on search results, a mix of user behavior and job similarity is considered. Under the job description, jobs are recommended based on how close the jobs are to the current job description that users are searching for. During my internship, I worked on a solution to address the “cold start” of new users who visited the platform but have yet to input any skills or resume details. Recommendations for this group of users are generated from search patterns of existing users who searched for the same jobs.
Recommender models are continuously improved to increase user exposure and engagement with job listings through a data driven methodology known as AB testing, a randomised experiment of two variants of a website. My role in the team was to improve the efficiency of conducting AB testing, through refactoring the codebase to allow for AB test to be launched and stopped with a simple API call, and wrote utility tools for the team to standardise the computation and comparison of performance metrics with hypothesis testing.
Besides the software related work, I also got the opportunity to suggest improvements to the model by analysing usage patterns on the MyCareersFuture website. I was tasked to find ways to improve click-through rates on job recommendations and job descriptions. With the specific information found in job postings that users click through such as salary, seniority, industry and occupation, I ran a logistic regression to identify users’ preferences and presented my case to the team who were willing to give my suggestions a shot.
Throughout my internship, I picked up useful tools such as PySpark, a big data framework, REST APIs with Flask and Docker to spin up a local development environment. I was exposed to various natural language processing algorithms and recommender model paradigms (user-based and item-based collaborative filtering) in production on the cloud. Other than learning about technical skills, my supervisors also provided great advice to how to improve my communication skills, specifically how to tie down the problem statement, and how to present your solution.
How it is like working from home
Despite working from home for the entire internship, I was still able to contribute effectively to the team through code reviews on GitHub’s pull requests and effective sharing of Jupyter notebooks on Databricks. My colleagues were very responsive, and I received timely feedback on my work through video calls and screen sharing.
While there were timelines to achieve targets for the team, there was much flexibility on how and when I wanted to complete intermediate goals. We used Slack extensively for daily standups to update our daily plans, and monthly meetings to consolidate progress and address any hurdles to each other’s work, if any. However, not all meetings were purely for work! Occasionally, we had team-bonding activities where we played games like scribbl.io and trivia to wind down and relax.
On top of opportunities for exposure to data science in GovTech, there were many online sharing sessions that we could attend, to learn about what the other teams were doing as well. I had the opportunity to attend brownbag sessions on frontend development, cybersecurity and application hosting by GovTech engineers in other divisions.
Final Words
Through this internship, I gained valuable experience on how a data science project runs and iterates in production, which is often not something you can learn in school. Through my conversations with my colleagues, the internship also gave me greater clarity for my career planning (Software Engineer? Data Scientist?).
If you are looking for a meaningful internship where you can make an impact on people’s lives through technology, there is no better place than GovTech! Apply to be an intern at DSAID; you won’t regret it.
