A Day in the Life of a Machine Learning Scientist at Integrate.ai - Garcia Liang

Reception at Integrate.ai

Reception at Integrate.ai

In March, we launched a new “A Day in the Life” interview series, to dig deeper into the day-to-day of professionals at the intersection of Analytics, Digital and Design. To kickoff this series, we were lucky to interview Garcia Liang, a Machine Learning Scientist from Integrate.ai.  

A Little on Garcia’s story

Garcia’s father is a professor in applied math, and despite Garcia’s initial lack of interest in statistics, he was forced to take his first Statistics course during undergrad. After that and without much convincing, he found a passion in this area and continued his study towards a PhD in Statistics at the University of Waterloo. He started his career as a Senior Data Scientist at Precima, the LoyaltyOne data analytics arm that specializes in customer analytics. After spending two years at Precima, Garcia joined Integrate.ai, an AI start-up as a Machine Learning Scientist.

Before we get into the details of Garcia’s typical day as a Machine Learning Scientist, we asked him a few burning questions from our readers!

(Garcia at his desk) Outside of complex data modelling and ML research, Garcia and many at Integrate.ai are board game fanatics!

(Garcia at his desk) Outside of complex data modelling and ML research, Garcia and many at Integrate.ai are board game fanatics!

What do you really do as a Machine Learning Scientist?

Garcia: My work as a Machine Learning Scientist’s consists of two parts:

  1. Client specific work and deliverables;

  2. Research and development, which includes conducting applied research and advancing the Integrate.ai platform with research-based and validated algorithms.

The first part requires understanding of the client business and the ability to translate complex concepts into layman-friendly and outcome-driven language. The second part is more in my wheelhouse, because it allows me to apply my academic background and experiment with new ML algorithms and techniques with a certain degree of freedom. I definitely have a knack for solving complex numerical  problems in creative and improved ways. At the same time, client-facing work is very rewarding! It’s always good to see the instant value of your work and get immediate client feedback.

I enjoy both of these job functions, so it’s important to me that I get a healthy balance between the two.

What are the differences between a Machine Learning Researcher and Machine Learning Engineer?

Garcia: Honestly, naming convention can be confusing sometimes.

Typically, Machine Learning Engineers are responsible for processing large volumes of data, building and implementing models. Machine Learning Scientists at Integrate.ai are similar to Data Scientists or Machine Learning Engineers in other organizations but they are also responsible for applied research. ML Scientists at Integrate.ai publish and contribute to research journals as well.

Intrusion in personal privacy is a common concern, do you use any personal data in your work?

Garcia: No, no, no !!! (No with three exclamation marks!)

Machine Learning Scientists do not have infinite access to people’s personal information. We are also not trying to build Skynet in Terminator! In fact, Machine Learning Scientists (at least in Integrate.ai) do not have access to Personally identifiable information (PII) data. All of the data is encrypted and stripped of any personal identifiers to avoid traceability back to individuals. Moreover, there are rules and restrictions that prevent any unintended data aggregation and cross-referencing.

Are you using some sort of special magic in developing algorithms?

Garcia: Not at all! Today, lots of ML applications are still employing algorithms that are not much different from traditional numerical methods previously used to identify data patterns and biases (at least for most use cases in the market).

Of course, with current advances in Machine Learning, we also experiment with more novel and complex models. So, to your question, it depends what you mean by magic.

Honestly, using simpler statistical models appropriately can, at times, be more powerful than using complex models, especially when you don’t have a deep understanding of the theory behind them. Some people still think Machine Learning is a black box. As a data professional and a researcher, we need to be diligent in uncovering the black box, and hone our ability to explain the inner mechanics and decision-making within the model to the general public.

Garcia’s Typical Day

Kickstarting the Morning

(Boardroom at Integrate.ai) A common place for client meetings

(Boardroom at Integrate.ai) A common place for client meetings

Garcia: For those familiar with agile projects, daily stand ups are not foreign. Folks at Integrate.ai follow a similar ritual to better coordinate cross-functional efforts between Engineering, Sales, Data Science and Machine Learning.

Of course, not every day is exactly alike. For the most part, I like to use my mornings to do some personal planning on the day’s activities and revisit any near-term research objectives. My mornings are normally filled with client-facing activities, which are highly collaborative, in contrast to my research-focused afternoons. We use the time to meet with clients for brainstorming, scope definition, solution design and demos.

Chugging Away in the Afternoon

Garcia: If time permits, I will try to spend my afternoons on research planning. Research planning can seem ambiguous so it’s important to define objectives that are outcome and application driven. When we come across client problems where there is no existing solution or if the current application is not optimal, we take the opportunity to turn these problem areas into research projects.

Concrete research deliverables and timelines are hard to plan since we often come across new learnings and breakthroughs along the way. To stay on track, my team and I put together a wish list of research-based algorithms and techniques that we want to try out and implement. We also go through a prioritization exercise to make sure our research reaches objectives defined at a quarterly basis.

I do most of my research work independently, run my own experiments, and continue to contribute to research journals and white papers. Sometimes, I spend time with my team to align activities for the next day.

Things can also get chaotic under tight deadlines, pressing client deliverables and competing priorities. In these situations, we spend more time with clients so we can solve problems more closely. Sometimes, ambition definitely comes with a price.

Integrate.ai follows a Software as a Service (SaaS) model, so there is added complexity to create models and platform capabilities that benefit all our clients. As part of continuous improvement and evolution of our platform, we also conduct experiments and solicit feedback from our clients and their users to validate whether something is working or not.

Tricks of the Trade

We asked Garcia to provide advice to any newcomers who are looking to pursue an exciting career as a Machine Learning Scientist. He also shared his personal views on how to survive in this overly competitive industry.

Getting Started as a Machine Learning Scientist

Garcia: I believe the following are pivotal in becoming a successful Machine Learning Scientist:

  1. Have solid knowledge and understanding of the underlying theory behind ML algorithms (You cannot call yourself a scientist if you don’t know what you are doing, right?)

  2. Hone your coding skills. You will need to roll up your sleeves and actually implement theoretical and research-based techniques in order to validate its feasibility.

  3. Get more experience in modelling and applied AI. Having the ability to build a robust algorithm is cool, but you also need to be able to translate business problems in modelling and ultimately help your clients and end users.

While it is not required to have a PhD to work as a Machine Learning Scientist, it is important to have the thirst for constant learning. And more than that, it is critical to apply this knowledge in real applications and use cases.

How to survive as a Machine Learning Scientist at a startup?

The toughest thing for me is learning to strike a balance between immediate impact versus high-potential research. - Garcia

Garcia: On one hand, there are tight deadlines and pressure to fulfill client or user needs and wants. On the other hand, there is the desire to work on meaningful research projects that will help the company scale—expand current capabilities and improve model performance. There is also a need to constantly work with your team and your clients to prioritize activities in order to maximize the value that you provide.

A few tips for folks who want to work as a Machine Learning Scientist in a startup environment:

  1. It is important that you are able to justify your research outcomes and be prepared to defend your point of view when challenged by others that are also invested in the project.

  2. Remember to see the bigger picture. As a service provider, we are not competing with our clients. Instead, we are open to integrate client solutions into our platform for mutual benefit.

  3. Keep in mind that clients also build great models and these models sometimes make more sense in their business context.

Last but not the least, it is imperative to take care of yourself! Startups have lots of overtime which is unavoidable in a rapidly growing work environment. It might take a toll on you and your family so make sure to try to find a balance whenever possible.

(Entrance at Integrate.ai) In the spirit of work-fun balance, Integaret.ai hosts monthly diversity socials for employees. Garcia specially enjoys the ritual of welcoming new recruits— new hires have half a minute (with no pause) to present a 20-slide introduction deck to the entire company!

(Entrance at Integrate.ai) In the spirit of work-fun balance, Integaret.ai hosts monthly diversity socials for employees. Garcia specially enjoys the ritual of welcoming new recruits— new hires have half a minute (with no pause) to present a 20-slide introduction deck to the entire company!

Parting Words of Advice

As the interview came to a close, we asked Garcia to provide some ending remarks on what keeps him passionate about Machine Learning. He told us that making people’s lives better is by far the most rewarding part of his job.

Take Amazon’s recommendation engine as an example. He personally does not find it invasive, but instead, extremely convenient and useful. This notion may cause some people to raise their brows in skepticism. But there is an undeniable convenience to Amazon’s recommendations and delivery service level.

However, as a Machine Learning Scientist who is working in the frontlines of applied AI in user-facing applications, Garcia emphasized the increasing importance of ethics. In recent years, more and more tech firms are endorsing the principles of ethical AI. In Garcia’s view, it is especially important for frontline AI professionals to embody these principles and embed these considerations into their AI solutions. Only with the continued efforts of data scientists, engineers and researchers, can we create a future where we can safely and freely reap the rewards of AI intelligence in every facet of our lives without having to worry about any intrusion to privacy or any ethical violations.

Featured Company

Integrate.ai is an AI startup in Toronto that helps traditional consumer enterprises become as customer-centric as Amazon. Integrate’s AI-powered software platform drives revenue growth with targeted machine learning applications and access to collective consumer intelligence across industries.

Do you work in the intersection of Analytics, Digital and Design and want to share your experience?

Talk to us! Share your viewpoints and help those new entrants seeking insight find some answers to their burning questions.

About the Author(s)

This article is co-written by Kitty Chio, the Content Lead at ABD, and Michelle Liu, the President of ABD.