Moscow, Russia

Ramil Gabdrakhimanov

ML Engineer at Yandex.

I build systems to evaluate and improve language models and AI agents.

LLM evaluation Reinforcement learning AI agents

Telegram Email GitHub

Experience

Yandex Current

Data Scientist / ML Engineer

LLM Evaluation, RL & Agents

  • Build benchmarks and evaluation harnesses for LLMs and tool-using agents.
  • Develop LLM judges, scoring rubrics, and programmatic verifiers.
  • Run reproducible model comparisons and investigate model, scoring, and pipeline failures.
  • Build SFT and RL data-generation pipelines and RL environments to develop models' agentic capabilities.
Technical details

I turn research benchmarks into reliable evaluation pipelines: examine task design and scoring, reproduce baselines, and check that results remain comparable across frameworks.

I evaluate agents on file, spreadsheet, and coding tasks across frontier and open-weight models. I analyze judge reliability, calibration, and multi-step failures, and aggregate results using rubric scores and model rankings.

I build pipelines that generate training data for SFT and RL, and develop RL environments to improve models' tool use and multi-step task solving. This includes tasks with programmatically checkable outcomes.

ecom.tech

Data Scientist · Previously

Formerly samokat.tech

Built search and recommendation systems for e-commerce, including personalized ranking, offline evaluation, A/B testing, and model monitoring.

Education

Central University

M.Sc. in Machine Learning

2025 — present

MIET

B.Sc. in Computer Science and Engineering

2022 — 2026 · Graduated

A little more

I enjoy turning research into working systems and learning by building things myself. Outside work, I like experimenting with hardware, and small technical projects — especially when they involve understanding how things work beyond the surface level.