At Kone Oyj, I developed an agentic framework to increase productivity in software development and ML experiments. I also optimized computer vision algorithms for edge devices and built predictive maintenance models to identify equipment faults before failure.
At Aalto University, I developed a multimodal pre-training framework for video retrieval and action localization, achieving state-of-the-art results on MSR-VTT and YouCook2. This work was published at LREC-COLING 2024 as a hierarchical cross-modal transformer with task-specific contrastive losses.

