Skip to content
View TrupologDS's full-sized avatar
😮‍💨
😮‍💨

Block or report TrupologDS

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
TrupologDS/README.md

Aleksandr Sorokin

Research Engineer / ML Engineer focused on LLM post-training workflows, model evaluation, and reliable production ML systems.

I build reproducible Python/PyTorch experiments, Hugging Face-based training and evaluation pipelines, retrieval systems, and production-oriented ML services. My portfolio is organized around the kind of work I want to do more of: post-training, evaluation, robust ML infrastructure, and practical systems that can be inspected, rerun, and monitored.

Featured Work

Project Why it is relevant
LLM Evaluation Harness Lightweight evaluation harness for behavior regression checks, prompt/output datasets, rubric-style metrics, error taxonomy, and generated evaluation reports.
Russian LLM Pretraining and SFT Experiment report and reproducible scaffold for tokenizer training, causal LM pretraining, LoRA SFT, model cards, and evaluation reporting.
Semantic Retrieval for arXiv Papers Dense retrieval pipeline with BGE embeddings, FAISS indexing, MRR@5 evaluation, latency profiling, and error analysis.
Multi-Task Information Extraction on NEREL PyTorch/Transformers multi-task model for NER plus document-level multi-label classification with threshold tuning and test analysis.
Text-to-Image Product Search with Fine-Tuned CLIP CLIP fine-tuning and embedding-index workflow for text-to-image catalog retrieval.
ML Engineering Projects Production ML portfolio covering Airflow, DVC, MLflow, FastAPI serving, monitoring, recommendation systems, and reproducible pipelines.

Focus Areas

  • LLM post-training: SFT workflows, LoRA adapters, tokenizer/data preparation, experiment reporting.
  • Model evaluation: task metrics, regression checks, error taxonomies, qualitative review loops.
  • Reliable ML systems: reproducible training, artifact hygiene, CI, typed Python modules, testable pipelines.
  • Production ML/MLOps: model serving, monitoring, orchestration, experiment tracking, and data validation.

Core Stack

Python, PyTorch, Hugging Face Transformers/Datasets, TRL, PEFT/LoRA, scikit-learn, pandas, NumPy, FAISS, MLflow, DVC, FastAPI, Docker, Airflow, SQL, GitHub Actions.

Contact

Pinned Loading

  1. llm-evaluation-harness llm-evaluation-harness Public

    Lightweight LLM behavior evaluation harness with prompt/output fixtures, metrics, error taxonomies, regression checks and generated reports.

    Python

  2. data-science-projects data-science-projects Public

    NLP, LLM, retrieval, computer vision and machine learning portfolio with reproducible notebooks, evaluation reports and project scaffolds.

    Jupyter Notebook

  3. ml-engineering-projects ml-engineering-projects Public

    Production-oriented ML engineering projects: pipelines, experiment tracking, serving, monitoring, recommenders and reproducible training.

    Jupyter Notebook

  4. data-engineering-projects data-engineering-projects Public

    Data engineering portfolio with SQL data marts, Airflow, Spark, Kafka, Docker/Kubernetes and warehouse pipelines.

    Python

  5. TrupologDS TrupologDS Public

    Profile README and portfolio landing page for LLM post-training, model evaluation and reliable production ML.