I'm a Software Engineer at Google DeepMind, based in Zürich, Switzerland. My work focuses on improving the Gemini App, where I worked on post-training for Factuality and Freshness (SFT, RL, enhancing retrieval) and now leading the development of the Classification Platform.
Previously, I developed large-scale solutions for Google Assistant and Google Cloud AI.
Throughout my career, I worked a lot with data and naturally adopted a data-driven decision-making approach. This has helped me develop a strong intuition for leveraging data to solve complex problems (primarily through different ML solutions like ranking, classification, clustering, etc.). Nowadays, I enjoy improving user-facing products by finding the optimal balance and tradeoffs between quality, latency, and resource consumption.
My immediate focus is on maximizing the utility of tiny encoder models: distilling the power of big models into lean, task-specific architectures for real-world deployment.