TOPICS

My main interest lies in the application of machine learning systems to real-world problems. Over the years, I have worked across domains such as bioacoustic modeling of dolphin communication, gesture recognition, web recommender systems, wearable computing, and online marketing.

Beyond machine learning, I enjoy exploring novel programming languages, databases, and networked systems. My industry work includes building job and contact recommendations at Xing, developing e-commerce and marketing tools at Shopify, and advancing audience building and ad targeting systems at Meta (Facebook). Currently, I develop machine learning solutions for logistics and fulfillment at Hive.

Research Background

My Ph.D. research focused on analyzing sequential data from biological signals such as dolphin vocalizations and human gestures. I developed data-driven methods for indexing, annotation, and discovery of high-dimensional sensor streams, aiming to uncover and model natural patterns within them.

Many forms of behavioral data — from animal communication to human activity — exhibit language-like structures. My research aims to automatically identify the atomic units of these structures and the temporal patterns of their occurrence. This approach enables interpretable analyses of complex continuous signals, such as understanding dolphin communication or improving gesture and activity recognition systems in everyday environments.

Methods & Techniques

Over time, I have worked with a broad range of modeling approaches, including deep learning, unsupervised feature learning, Hidden Markov Models, and grammar induction through alignment-based learning. Depending on the problem domain, I have applied these methods individually or in combination to analyze sequential data, discover structure, or build recognition systems. My toolbox also includes clustering and indexing algorithms such as k-means, spectral clustering, hierarchical clustering, and indexing techniques like iSAX and HNSW. For sequential and bioacoustic data, I often use alignment algorithms such as Dynamic Time Warping (DTW), Smith–Waterman, and Needleman–Wunsch. In applied industry settings, my work frequently involves ensemble methods such as Random Forests and XGBoost, along with experimentation frameworks for A/B testing and causal inference to evaluate and interpret the real-world impact of machine learning systems.

Vision

In general, I aim to create intelligent agents that discover how data is naturally organized and communicate those structures in human-understandable form.