ML systems builder working on evaluation, robustness, and real-time systems.
RU / EN / ES
I build technical artifacts around:
- agent evaluation;
- safety under distribution shift;
- Android trust-state measurement;
- repo-context tooling;
- real-time ML systems.
I care about runnable artifacts, reproducibility, failure modes, measurable tradeoffs, and systems that can be inspected instead of only described.
The common thread across these projects is reliability under messy conditions:
- distribution shift;
- unsafe overconfidence;
- trust boundaries;
- noisy real-time systems;
- evaluator validity;
- reproducibility;
- context quality for coding agents.
My main strength is turning ambiguous technical problems into runnable artifacts with tests, generated outputs, and clearly stated limitations.