From Physics to Kaggle Grandmaster: A Data Scientist’s Journey and Advice

Physicist Bojan Tunguz, born in Sarajevo and now a data‑science professional at H2O.ai, holds a PhD in physics from the University of Illinois and a master’s in applied physics from Stanford. After finding academic positions scarce—especially as an immigrant—he turned to data science and Kaggle competitions, which offered the right balance between science and technology.

Q: How did you transition from academia to industry?

A: Academic jobs are hard to find, especially in saturated fields like theoretical physics, and being an immigrant makes it ten times harder. I first tried tech writing and building small desktop PCs, but data science captured me because it bridges science and technology perfectly.

Q: How difficult is it to become a Kaggle Grandmaster, and what attracted you to Kaggle?

A: I am a Grandmaster in kernels and discussion competitions and hold a personal gold medal. Each category demands different skills: competitions require top‑level technical ability, while kernels and discussions also need communication, writing, and data‑visualisation skills. Understanding competition dynamics—what is posted and when, and what the community finds valuable—is crucial.

My first gold medal came from an IEEE camera‑sensor recognition competition about a year and a half ago, and a year ago I won a housing‑credit‑default‑risk competition, which remain my most memorable achievements.

Q: How do you decide which competitions to enter?

A: I simply input all of them and usually pick one or two. I consider how many modeling pipelines are already built, the effort needed for a solid baseline, the stability between local scores and leaderboard rankings, and the potential ensemble gain.

Q: What is your typical workflow for a Kaggle problem? Any favourite ML tools?

A: I start with small experiments and gradually scale up, never chasing the “best” solution immediately. My standard toolbox includes pandas, numpy, scikit‑learn, XGBoost, LightGBM, and Keras.

Q: As an H2O.ai data scientist, what is your role and focus area?

A: I mainly help customers solve their data‑science and ML problems. I also contribute to autonomous AI, focusing on feature‑transformation “recipes” and unsupervised learning.

Q: If you could collaborate with any H2O.ai masters, who would you choose and why?

A: Everyone! Before joining H2O I already collaborated with Olivier Gralier, and now I work with Dmitry Larko. Many H2O.ai colleagues are people I have admired for years, and it’s an incredible honour to call them teammates.

Q: What is the best thing you learned from Kaggle that you apply at H2O.ai?

A: One of the most important “meta” skills is the ability to adapt quickly and try many different approaches. Because most great ideas have a chance of failing, you should always have at least one contingency plan, preferably six.

Q: How do you stay up‑to‑date with rapid developments in data science?

A: Competing on Kaggle lets me apply at least one new technique, library, or framework for each competition. It is arguably the best way to keep ML application skills sharp and current.

Q: Do you wish to apply your expertise in specific ML domains?

A: I love NLP and hope to spend more time on it, as it aligns with classic AI problems. I’m also interested in fintech, though it remains more complex than pure science. Recently I performed well on “pure” science problems ranging from protein classification to predicting scalar coupling constants using only ML.

Q: What advice would you give to aspiring data‑science hobbyists?

A: Now is the best time to start a data‑science career. The field is extremely open, with abundant resources: textbooks, blogs, webinars, Kaggle, MOOCs, forums, open‑source tools, and access to practitioners. Build a reasonable plan that fits you, develop both technical and soft skills (writing, communication, networking), be patient, learn from failures, and enjoy the journey. Becoming a Kaggle Grandmaster requires massive effort, persistence, and focus, but with the right tools you can direct your effort effectively.