Designed and developed a Transformer based foundational multimodal user behavioral model to detect different attack vectors across product areas. Leading technical efforts to prevent abuse at scale by developing near-real time detection infrastructure at 250k QPS that detects ~XM abusive accounts / day. Developed new Ad product to increase revenue by 0.XB ARR.

Reduced by 20% the CPU consumption of a 120k QPS real-time detection system. Developed and launched multiple Machine learning models thatdetect ~XXk hijackers / day.

Developed Python software to simulate beam stabilization effects in the LHC, leading to a x30 speedup in simulation time: faster simulation time means more trials can be run on the same hardware.

Deployed quantized models on test cars using Qualcomm open source software, leading to 10% reduction in model size. Developed novel computer vision architecture using PyTorch to detect smaller objects in spite of larger ones.

Engineering new domain specific language to express epidemiology models concisely, reducing coding lines by 70%.

I have created the open source library Surrogates.jl

BSc + MSc in Applied Mathematics