Applied mathematician building ML systems.
AI researcher at VK — distilling SOTA encoder transformers into state-space architectures. Kaggle Competitions Expert. Founder of dot. and LooksAR. Applied Mathematics at HSE University (BSc ’28).
now
Knowledge distillation of encoder transformers (ModernBERT, Qwen Embedding) into state-space and efficient-attention architectures at VK, supervised by the author of DenseAttention (ICML 2025).
selected work
Exoplanet spectra extracted from noisy telescope time series with Gaussian processes, Bayesian inference and physics-informed modeling — no neural nets.
Combinatorial optimization with custom Rust and C++ packing solvers and genetic algorithms.
AI planner that turns scattered thoughts into calendar plans. Production iOS app, payments infrastructure, MCP integration — any model via API token.
AI virtual try-on widget for optical e-commerce: 3-second facial measurement, Fit Score, embeddable into product pages.
Per-category account price prediction for a large marketplace, deployed via ONNX.
also
writing
Writeups in the works: the Ariel solution (why Gaussian processes beat neural nets here) and an interactive note on GP denoising.
contact