Skip to content
View BearAx's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report BearAx

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
BearAx/README.md

Aleksandr Medvedev

ML Research Engineer / ML Systems Engineer
Model Quality · Efficient Training · Evaluation · PyTorch Systems

I improve AI models where quality meets systems: reproducible experiments, controlled ablations, robust evaluation, and training/inference optimization for speed, memory, and cost.

LinkedIn · CV


Operating principle

If I claim an ML improvement, it should have a baseline, an ablation, a metric, and a cost.

I am focused on practical AI improvement: making models better, more stable, faster, and cheaper to run through disciplined experimentation and systems-aware engineering.


What I work on

Model Quality training recipes, fine-tuning, robustness, calibration, data-centric improvements
ML Systems Efficiency profiling, AMP, torch.compile, batching, checkpointing, latency, throughput, memory
Research Engineering baselines, ablations, multi-seed evaluation, tracked configs, reproducible reports

Current portfolio direction

Project Focus What it demonstrates
ml-systems-lab Training/inference efficiency Profiling PyTorch workloads, measuring latency/throughput/memory, reducing cost with AMP, compile, batching, checkpointing
vision-recipe-bench Model quality through training recipes Controlled ablations for optimizer, LR schedule, augmentation, EMA, regularization, robustness, calibration
small-lm-lab Small Transformer LM training Tokenization, sequence packing, perplexity, training loop discipline, efficiency-quality trade-offs
nlp-ft-discipline Fine-tuning stability Seed variance, calibration, validation hygiene, robust evaluation for Transformer classifiers

Tech stack

Core ML: Python, PyTorch, Transformers, CNNs, small LMs
Experimentation: W&B / MLflow, Hydra / config-driven runs, ablations, multi-seed evaluation
Efficiency: CUDA/NVIDIA GPUs, AMP, torch.compile, profiling, checkpointing, batching
Engineering: Linux, Docker, Git, GitHub Actions, reproducible pipelines


What I am building toward

I am aiming for roles where I can work on the practical side of improving AI systems:

  • ML Research Engineer Intern
  • ML Systems Engineer Intern
  • Applied ML / LLM Engineer Intern
  • Model Quality / Evaluation Intern

My preferred work is at the intersection of:

better models + reliable experiments + efficient training/inference

Repository standards

For serious ML projects, I try to include:

  • train.py, eval.py, configs/, scripts/
  • fixed seeds and reproducible configs
  • baseline + ablation table
  • training curves and metric plots
  • latency / throughput / memory measurements when relevant
  • results.md with what worked, what failed, and what I would try next

GitHub activity

Pinned Loading

  1. Lambda-Parse-and-Play Lambda-Parse-and-Play Public

    Haskell 1

  2. Distributed-Text-Mining-and-Sentiment-Analysis Distributed-Text-Mining-and-Sentiment-Analysis Public

    Project for Distributed Network Programming

    Python 2