cv

Research Fellow

Goodfire AI

• Research project on mechanistic interpretability. Developed the neuron community discovery method to identify neuron-sparse representations in LLMs.

AI Safety Fellow

Anthropic

• Research project on capability removal. Developed Selective Gradient Masking, a pretraining intervention to localize target knowledge in LLMs for later removal.

Staff Software Engineer

Meta

• PyTorch framework for differential privacy. Lead developer and maintainer of Opacus, a PyTorch library for differentially private model training, used by Meta teams and external researchers (2,000+ GitHub stars).
• Privacy-preserving ML. As a research engineer, applied differential privacy and federated learning to products and supported research on differential privacy and privacy attacks.
• StopNCII.org. Led development of the privacy-preserving platform with the UK-based NGO Revenge Porn Helpline, using on-device perceptual hashing to help combat non-consensual intimate image sharing.
• Safety ML. Led a team of four engineers building models to detect and remove harmful content.

Senior Software Engineer

Zvooq

• Recommender system. Led a team of four engineers building a music recommendation engine.
• Analytics engine. Built an internal analytics system for copyright-holder reporting.

Software Engineer

Mail.ru

• Search ranking. Improved the ranking algorithm for Mail.ru's search engine, which held roughly 10% of the Russian market at the time.

Software Engineer

Cốc Cốc

• Search infrastructure. Built a distributed web crawler and improved query intent classification.

PhD student

Imperial College London

Computing Research

Specialist degree (5 years)

Lomonosov Moscow State University

Applied Mathematics and Computer Science