Irina Belousova

I am Tech lead/Senior ML manager on team focused on LLM Optimization. I am designing and implementing ML-based solutions to improve runtime latency, training time, memory usage, time to first token, and decoding speed across all Apple applications. My top strengths are the ability to formulate research problems, design, experiment, implement, communicate, and integrate solutions into products. I excel at innovating across the entire end-to-end ML production pipeline, including dataset creation, neural network architecture design, model training, fine-tuning methods, training time optimization, and both on-device and server-side inference. Previously, I was an ML Engineering Manager and Technical Leader of the Natural Language Representations team, which powers natural language understanding in Siri for both server and on-device applications. My team uses a range of natural language processing and machine learning technologies. We developed NL embeddings and a set of model compression techniques that power many different applications, including Siri, Photo Search, Mail, Calendar, Spotlight, Safari, and Generative Playgrounds across all Apple devices.