Data selection for ASR. I was able to train models on 10% less data and achieve the same accuracy reducing transcription costs and reducing training times.

Few-shot learning for action recognition with ego-centric videos.

Part-time intern working on applying ideas in Continual Learning to Self-Supervised Automatic Speech Recognition, https://arxiv.org/abs/2107.13530.

Mining the ghtorrent database to quantitatively understand what makes a leader, preprint: https://www.research-collection.ethz.ch/bitstream/handle/20.500.11850/463810/1/CLE_WP_2021_01.pdf.

Worked as a data scientist, digital consultant and web developer, doing cool work like helping ship operators predict when an engine fault will occur to helping Londoners plan their journeys and making their journeys less painful.Clients included: TfL, Vodafone, EE, BAE Systems & Network Rail.

I wrote a tool in C++ and Fortran 95(!) to help researchers characterize plasma jets in the lab. The curve-fitting tool was able to compare experimental data, to theoretical models to interpolate parameters of the plasmas produced in the lab based on these theoretical models.