The Minds Behind the Models: Kei Eugene Kwok

Kei Eugene Kwok: Systematic Feature Narrowing from the Heart of Hong Kong's Quant Scene

Kei Eugene Kwok brings a formidable academic and professional pedigree to the competition. He holds a Master of Advanced Studies in Applied Mathematics from the University of Cambridge (First Class Honours), and a Bachelor of Science in Mathematics from UCLA, where he graduated Summa Cum Laude with a perfect 4.0 major GPA. His journey began even earlier at Westminster School in London, where he excelled in the British Mathematical Olympiad.

Today, he works in Hong Kong as part of Prime Strat at Goldman Sachs. Before that, he was an Algo Execution Quant at CLSA, focusing on predictive modelling and Transaction Cost Analysis (TCA).

With that kind of background, you might expect his approach to AlphaNova to involve highly exotic, proprietary models. Instead, his entry point into quant finance was remarkably grounded.

"I was drawn in through a mix of academic curiosity and hands-on exposure. During my mathematics studies, I gained exposure into quant finance as the intersection of probability, statistics, and real-world decision-making. AlphaNova caught my attention as a way to test some of those ideas."

Prediction Markets and the Flow of Information

When asked about a recent project that excites him, Kei pointed to a fascinating corner of the financial world that is currently exploding in popularity: prediction markets.

"I've been actively exploring prediction markets—particularly how flow structure there compares to traditional financial markets. I'm currently trying to script up something in Python to better understand the data and market mechanics."

Comparing the flow structure of prediction markets (like Polymarket or Kalshi) to traditional equity or futures markets is a rich research area. In traditional markets, order flow is dominated by algorithmic execution and market makers. In prediction markets, the flow is often driven by retail sentiment and event-driven information asymmetry. Bridging that gap requires a deep understanding of market microstructure—exactly the kind of work he does at Goldman Sachs.

The Systematic Playbook: Expand, Narrow, Ensemble

For the AlphaNova competition, Kei relied on a disciplined, iterative process. Rather than jumping straight to the most complex neural network, he started broad and let the statistics do the pruning.

"I take a systematic approach: start with a broad set of expanded features, then iteratively narrow down using statistical checks. I typically test a few model families—linear and tree-based models, and tune each with cross-validation. I also experiment with ensembling to improve robustness."

It is a classic quant workflow, and it echoes the sentiment of other top performers in this series. Utkarsh spoke about starting with one clean idea and making it robust. Alok leaned heavily on Ridge regression on cross-sectional ranks.

Kei’s approach of expanding features only to ruthlessly narrow them down is a direct defense against the noise that dominates financial data. By relying on linear and tree-based models with cross-validation, and then ensembling them, he prioritizes stability over chasing a fleeting, overfit signal.

A Nuanced Take on AI: Automation vs. Analytical Judgment

Kei uses AI extensively, but his perspective on its limits is one of the most nuanced we've featured.

"I've been using LLM-based coding assistants extensively—especially for rapid prototyping, refactoring, and debugging. They've drastically cut down time spent on boilerplate code and syntax errors, freeing me up to focus on model design and analysis. At work, I've integrated them into my daily workflow with Copilot, which has been a game-changer for productivity."

Like many researchers we've spoken to, AI has removed the friction between having an idea and testing it. But when asked about how he uses AI in his workflow, he drew a very clear line between execution and judgment.

"I use AI tools primarily for automation—writing boilerplate, debugging, and exploratory data analysis... That said, I find AI less helpful for tasks that require true creativity, analytical judgement, and organisational reasoning all into one—those are areas where I lean more on my own thinking and rely more on my objective judgment."

This is an important counterpoint to the narrative that AI is taking over the entire research loop. For Kei, AI is a highly capable junior engineer. It writes the code and helps explore the data. But the decision of what to build, which features to keep, and how to interpret the results remains firmly human.

It is the same boundary Mathurin enforced when he noted that language models are exceptionally good at producing plausible, convincing narratives for noisy results. The ultimate filter must be the researcher's objective judgment.

What We Can All Learn

Kei’s profile offers three grounded lessons for anyone building systematic strategies:

Start broad, then narrow ruthlessly. Expanding your feature set is easy. Using statistical checks to cut it back down is where the real work happens.
Rely on proven model families. Simple linear and tree-based models, properly cross-validated and ensembled, will often beat complex architectures that are harder to debug and prone to overfitting.
Let AI handle the syntax, keep the judgment for yourself. Use tools like Copilot to accelerate your workflow, but recognize that true analytical reasoning and creativity cannot be outsourced.

That combination of elite technical execution and grounded analytical humility is exactly what the AlphaNova community is built on.

Stay tuned for more profiles from the AlphaNova community. If you’d like to be featured, reach out—we’d love to share your story.