A Day in the Life of a Principal Data Engineer at MOHT

Although I have been serving as Principal Data Engineer at MOHT since February 2021, I was actually trained in electrical engineering during my undergraduate studies. I was “accidentally” introduced to the fascinating world of neural networks by my FYP supervisor Professor Yen Hsu-Chun. I vividly recall the first Hopfield networks paper he recommended me to read. Thereafter, I went all-in, studying artificial neural networks in the 1990’s in windy Hawaii, and eventually completed a Ph.D. thesis in Texas on non-linear dimensionality reduction using manifolds in Texas.

My career spans start-ups, tech giants and the academia. After graduating in 2000, I worked at a US start-up. Post 9-11, in 2003, I left the US for Singapore to teach computer engineering at NTU until 2013. My research interest was sentiment analysis and information retrieval or search engines.

In 2016, I joined WiFi MasterKey (WiFi), where, as Chief Scientist, I was fortunate to work on terabytes of daily incoming data long before companies like Shopee, Tencent, Alibaba, ByteDance set up their operations in Singapore.

My first foray into healthcare was at Asus’ AI division. As senior engineering manager, I worked on projects at the intersection of healthcare and AI, such as an app that routes outpatients to different specialists based on their symptoms, and MIA (Medical Intelligence Access) – a PubMed search engine.

In November 2020, I met Prof Robert Morris, Chief Technology Strategist at MOHT. MOHT was a 2-year-old organisation then, and doing work that is a confluence of all the things I enjoy: data science and tech, innovation, healthcare. I was very inspired and impressed by Prof Robert’s energy and passion for healthcare. Like me, he has a background in tech, and is at the stage of his career and life where he wants to give back to society, by harnessing his skills and knowledge in data science and technology to improve the way health is delivered to patients. He was able to articulate the development, the mission and purpose of each project he oversees. One project stood out for me: mindline.sg. mindline.sg is a web-app platform providing validated self-assessment, evidence-informed self-care exercises, AI Chatbot, and peer-support forum.

Since joining MOHT in February 2021, I’ve been a part of the data science and technology team and working on mindline.sg. I helped set up a data platform called MDP (MOHT Data Platform) which facilitates real-time analysis of data from mindline.sg. Prior to MDP, data analysis was done by individual data scientist or analyst using excel or python charting, with csv file exchange as the most common way of data sharing. With MDP, dataset and analytical results can be easily shared among various teams and colleagues. Most importantly, MDP maintains just one copy of the database, instead of the classical two stage approach.
As healthcare was an unplanned career move, it is a relatively new domain for me. While I have been mostly focused on the technological aspect of healthcare, I now have an opportunity to contribute to digital mental health, working on a system that facilitates the mental health and wellness of youths, working adults, and the general population.

Recently, I have been working with the Integrated Health Promotion team on their Digital Local Connect initiative, a one-stop resource for community volunteers and service providers (general practitioners) to find relevant services and activities to recommend to residents during face-to-face visits.

I like the variety of tasks I get to do daily, ranging from app development (mainly python) to SQL data and visual analysis to gain deeper insights into the mindline.sg data, to updating the ETL (Extract Transform Load) scripts that runs daily to clean and extract mindline.sg backend data. From time to time, I do some DevOps tasks such as patching a module or updating a component on the backend Kubernetes system that is serving MDP.

Any of the above tasks could require several hours or days, especially when conducting deep data analysis. Sometimes, we must model the data distribution, at several different combinations of dimensions, before we can find the cause. This is extremely enjoyable as I get a Eureka moment once I find the source.