A Day in the Life of Helen Ngo, Machine Learning Engineer at Dessa
Continuing our much-anticipated “A Day in the Life” interview series, we identified more top-tier practitioners and continued to uncover the story of professionals at the intersection of Analytics, Digital and Design. This month, we had the pleasure to chat with Helen Ngo, a Machine Learning Engineer from Dessa.
A Little on Helen’s Story
Helen graduated from Western University with a specialization in mathematics and a certificate in writing. Fresh out of school, she wondered what she could do with a background in pure mathematics and numerical methods. Coincidently around this time, Toronto was quickly becoming known as the epicentre of machine learning and deep learning research. While spending some years building her career in Data Science, Helen discovered machine learning and was eager to tap into its huge potential in real-world applications. She felt that she can be part of this world-class talent force and was determined to join one of Toronto’s most innovative startups in the field of AI.
Soon after, she joined an enterprise in Telecommunications and contributed in building machine learning models that powered the call centres, retail stores and web applications. Through the Deep Learning in Production meetup hosted by Dessa, Helen came to learn about the firm’s diverse projects and decided to join the flourishing AI startup early this year.
Helen is very active in the Data Science and Machine Learning community and is extremely passionate about bringing people together to share knowledge and new ideas. She was part of the Toronto Deep Learning Series (now Aggregate Intellect’s Socratic Circles) Steering Committee and currently organizes the Toronto Women’s Data Group. She also furthered her interest in writing by being a part of the editorial staff at Towards Data Science. Most recently, Helen was selected as one of the 12 early-career professionals to join the Sidewalk Toronto Fellows Program, a fellowship that garner research recommendations from multi-disciplinary perspectives in regards to the use of urban design and digital technology at the Quayside site. Fellows formulated their research based on study tours and will publish findings and recommendations that directly influence the Sidewalk Toronto Master Innovation and Development Plan (MIDP).
Machine Learning Engineer @ Dessa
Can you describe what you do at Dessa as a Machine Learning Engineer?
Helen: Dessa helps enterprise teams build and ship artificial intelligence applications from ideation, prototyping to production. My role as a Machine Learning Engineer is split into two main parts. One of them is working directly onsite with clients, helping them build AI solutions that solve real business problems. The other main area is researching and developing tools that streamline Dessa’s machine learning development and production processes.
We work across various different industries from Telecom to Insurance, so each engagement is as interesting as the next. Typically, we integrate deeply within the client’s data science team. Not only does this allow us to collaborate closely in delivering the solution, it also allows us to clearly understand the learning outcomes that they are trying to achieve, and recommend next steps that would help them achieve these goals. An engagement team usually consists of Machine Learning Engineers, an Integrations Engineer and an Engagement Lead who helps bring together all of the project's various business and technical stakeholders. Depending on infrastructure requirements, we pull in other internal experts as needed.
What would you say the main differences are between working on machine learning in enterprise and working as a Machine Learning Engineer at Dessa?
Helen: This is a great question because job descriptions for these roles are not yet standardized across different companies. In general, the biggest difference I’ve noticed so far is in the scope of the machine learning projects I’ve worked on. As a Machine Learning Engineer at Dessa, I’m collaborating with clients on all parts of the machine learning workflow, from ideation to production. This wouldn’t necessarily be the case in enterprise, where roles tend to be more specialized.
At Dessa, we also have the unique opportunity to pick up business knowledge across various industries. We learn a lot about different industries by partnering up with data and ML teams that are part of our clients’ organizations. Right now, I am working with a large financial institution in the US. Though it’s my first exposure to working on ML for financial services, I’ve been able to pick up the necessary domain knowledge quickly.
What are some common gaps in capabilities you’ve seen in your work with clients?
Helen: At Dessa, we focus on enabling enterprise AI systems to run in production and ensuring that the implementation has a real-world impact on the business. From what I’ve seen, they typically have extensive experience with proof-of-concepts and research outcomes. The gap is in deploying trained models into production, and translating research into day-to-day business impact. This is especially hard to do at a large scale.
One of the biggest hurdles with this is figuring out how to seamlessly integrate our machine learning solutions into complex enterprise architecture and infrastructure, some of which has been around for decades. In many cases, the client has already built great data foundations--but often not with machine learning in mind. Great data engineering experts are able to dive in and wrangle data into the format necessary for machine learning, but the bigger problem is when there are required changes in an enterprise’s technology stack. These giant systems design problems are very much like solving a puzzle, and I think it’s fun to bring all the pieces together.
Helen’s Typical Day
Helen: I like to go over new research publications before I get into work, and definitely make sure to check out Jack Clark’s Import AI newsletter every week, which has a good overview of the latest news from research and industry. For articles that are a little bit less academic, I find Towards Data Science to be a great source. Engineering blogs from tech companies like Airbnb, Uber, and Stitchfix are also very informative, especially when it comes to developing real-world ML at scale. We also have a company-wide Slack channel where everyone shares interesting reads regularly.
The machine learning team also have other “mob-reading” initiatives, like our weekly “arXiv Sanity” sessions where we deep dive into trending ML papers that are interesting. The wealth of diverse technical experience at the company means that these sessions are always really productive. Machine learning is now such a large space and putting it all together can require specialized knowledge in certain subfields. Luckily, we have the aggregate knowledge to make this possible.
Pair Programming and Translating Research into Real-Life Value
Helen: I also spend time translating the latest research into new applications for industry.
We’re big proponents of pair programming at Dessa. It is extremely insightful to understand how another engineer approaches a problem. We learn the best of each other’s coding tips and tricks in the process. At times, we might involve more engineers and host a mob session to deep dive on a particularly challenging problem.
We also appreciate when members of the wider machine learning community share their code along with the release of their papers. This open-source sharing makes it easier for us to experiment and apply algorithms in actual implementations, as well as reproducing key results from the literature.
Outside of Working Hours
Helen: Being very involved in Toronto’s machine learning community, you can often find me at community events after work. Dessa values the machine learning community here in Toronto, and we often host Meetup groups and community events at our headquarters.
On top of that, the company encourages us to lead our own personal projects with company support provided. We are able to devote a portion of our time to work on these projects during work hours—kind of like the 20% policy at Google. Mine is currently top secret so I can’t elaborate on it! But a great example of such projects is one that my teammate recently released, called space2vec. They’ve created a model which is the best deep learning system in the world right now for detecting supernovas.
Outside of all the learning and self-guided projects, we also have a lot of fun as a team at Dessa. Last month, we had a fun internal event to celebrate Pi Day. We also regularly have fun team events like playing role-playing games or going rock climbing.
Get Started as a Machine Learning Engineer
Helen: I believe what makes people successful in this field is the ability to adapt and pick up new skills quickly.
Machine learning models are often Python-based, but there are entire ecosystems and support infrastructure around these models which are never quite the same across companies. Being able to adapt and quickly contribute in new environments is invaluable. It is impossible to know every facet of machine learning and infrastructure all the time, so being able to pick up what you immediately need to learn, to grasp and apply concepts quickly, and to prioritize a direct path to deliver on a project are all extremely important skills for success.
Is a PhD or a strong background in mathematics or software engineering required to break into Machine Learning?
Helen: I think it really depends on what you want to do. Many people at Dessa have come from a mathematics background and picked up the software engineering and coding skill sets afterward, but we also have people with a core background in software engineering and computer science.
When it comes to applying ML in the real world, I don’t think a PhD is necessarily a requirement. I’ve found that while a graduate education may get you in the door initially, I don’t think it’s a make-or-break factor. That said, at Dessa we integrate with the academic community and are required to understand research publications, so a strong mathematical foundation for these areas is crucial.
Other than a strong mathematics background, software engineering capabilities are also a huge asset. The ML industry right now is similar to the software engineering world a few years ago. We are facing similar challenges in the standardization of tooling and workflows as seen in the software engineering industry previously. For example, versions of Tensorflow are changing so quickly that code can constantly become out-of-date. And the ML workflow still varies quite drastically across different organizations and individuals. To make things more standardized, we’re looking to build upon the best practices learned from the software engineering world to ML.
What expertise or skills do you look for in a Machine Learning Engineer at Dessa?
Helen: Strong coding skills and a mathematics foundations to understand algorithms really help. One of the key artifacts we look at is a candidate’s portfolio of projects, especially if a candidate doesn’t yet have industry experience. It’s important to demonstrate that you are able to build something concrete from an idea and execute a project end-to-end.
A big part of what we do here is building and integrating ML algorithms into businesses and real-world applications. We like to see that the candidate is able to formulate real problems and apply the appropriate machine learning algorithms. Problems in real business settings are often messy and unclear. Creativity and experimentation are key, because these problems almost never fit in the nice boxes that we see in the research lab or Kaggle competitions.
The Highlight and the Struggle
Helen: On the flip side, it is becoming increasingly tough to manage my time. It might seem a little extreme, but I run my life on Trello! I constantly have things that I want to learn or create, so whenever something pops up, I add it to my backlog. There was a point where I tried to learn everything but, of course, that didn’t work out. Time is so precious—it is important to set immediate and long-term goals and focus on what is most relevant to you.
What I am most excited about is the potential impact that machine learning applications have on lives at a gigantic scale. This is a main motivator for me, especially in this role at Dessa. Our projects touch millions of lives daily through consumers across multiple industries all over North America.
I believe in the notion of being “all in” on whatever I choose to do, so I’m passionate and driven to constantly learn and contribute in this space. Of course, being in Toronto means that it’s hard not to be excited about AI right now--especially in such a rapidly growing startup environment. This is definitely a shared sentiment across the entire team of Machine Learning Engineers working today at Dessa!
Dessa collaborates with the world’s largest and most complex organizations to build real-world value with AI. Since 2016, the Dessa team has translated the latest AI research into impactful applications for industries ranging from finance to telecommunications. Committed to engineering excellence, Dessa has a diverse team of world-class experts in applied Machine Learning and software development.
Want to learn more about the careers in Data Science and AI?
Sign up with your email address to receive news and updates on our latest articles. Read about Garcia’s career at Integrate.ai if you missed it last month!
Do you work in the intersection of Analytics, Digital and Design and want to share your experience? Talk to us! Share your advice and help new entrants find answers to their burning questions.