Skip to content
View denis-samatov's full-sized avatar
:electron:
:electron:

Highlights

  • Pro

Block or report denis-samatov

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.

Content in all repositories owned by your account will be closed.
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
denis-samatov/README.md
Denis Samatov — Machine Learning Engineer & Technical Lead

Machine Learning Engineer & Technical Lead · LLM/RAG systems & evaluation · Medical imaging · Scientific ML

I build ML systems that can be checked: LLM/RAG assistants with release-blocking evaluation, medical image segmentation with patient-level validation, and scientific ML research software with documented reproducibility.

CV (PDF) LinkedIn ORCID Google Scholar


🧩 Public projects

Project Area What the repository shows
RuMed ICD benchmark Medical NLP / LLM evaluation Public Russian ICD-10 benchmark: TF-IDF vs zero/few-shot prompting vs retrieval, 822 test cases, bootstrap intervals and recorded token usage. LoRA training and vLLM serving results are pending.
TBMD Scientific ML Tensor-based modal decomposition and sparse sensor placement library; nested leave-one-scenario-out Brugge benchmark; reproducibility guide including an audit of the superseded arXiv v1 results
NAB / nab-bayes Bayesian inference Installable Python wheel and source archive for MCMC/Gibbs appraisal over Voronoi cells; development repository is private.
Yandex Workspace MCP Agent tooling MCP server for Yandex Disk & Wiki: 50+ typed tools, read-only by default, permission gates, audit logging, contract tests
LLM Adviser — case study LLM / RAG architecture Sanitized architecture of an engineering-knowledge platform in development (team of 5): deterministic ingestion, provenance, traceable retrieval
mcp-capguard Agent tooling pytest plugin that asserts which tools each MCP configuration profile exposes
Agentic architectures LLM agents 23 educational LangChain/LangGraph agent patterns, grouped by the failure each one addresses
OCR + LLM document pipeline Document AI Docling / RapidOCR / Ollama pipeline on a synthetic fixture

🔒 Selected private work

Source code is private (institutional or client work); code walkthroughs are available to hiring teams on request. The summaries below are author-reported; public repositories do not independently establish clinical validation or operational deployment.

EPIFAT — state-registered cardiac CT software (No. 2025610317): Attention U-Net pericardium segmentation, reported Dice 0.91 in a single run with patient-level splitting; GUI + CLI, manual correction and radiomics. Training data and held-out predictions are private; this is a research result, not a clinical qualification claim.

CourseLLM — source-grounded RAG assistant in use; golden-QA release gate, streaming/queued API, concurrency control. Team project I lead.

Housing knowledge-base RAG assistant — in use; hybrid BM25 + dense retrieval with reranking, 37-query retrieval regression (Hit@5 1.00, MRR@10 0.89), RAGAS release gate. These scores describe the fixed regression set; broader retrieval quality and operating scale require separate evidence.


🔀 Merged upstream contributions


💼 Experience

  • Machine Learning Engineer / Technical Lead (team of 3–4), Analytics and Machine Learning Department, MSUU — Apr 2024 – Present
  • Data Scientist / Research ML Engineer (part-time; leads a 5-engineer LLM team), Heriot-Watt TPU Center — Sep 2024 – Present
  • Machine Learning Engineer, Cardiology Research Institute — Mar 2021 – May 2026

🔬 Research

  • arXiv:2607.09687 — Tensor-Based Modal Decomposition and Sparse Sensor Placement for the Brugge Field Simulation Model (v1; revised manuscript with nested validation in preparation).
  • Talks (2026): oral presentations at AI4X Accelerate (Singapore) and the 6th International Workshop on Mathematical Geophysics; presentation at Data Intelligence in the Oil and Gas Industry (Nizhny Novgorod); ePoster at the SPE Annual Caspian Technical Conference.
  • Peer-reviewed cardiac MRI radiomics studies: Digital Diagnostics, 2024; Russian Journal of Cardiology, 2026; Siberian Journal of Clinical and Experimental Medicine, 2026.

⚙️ Stack

Python PyTorch scikit-learn LangChain LangGraph LlamaIndex MCP Ollama GraphRAG BM25 RAGAS Qdrant pgvector PostgreSQL Redis FastAPI Docker GitHub Actions Prometheus PyRadiomics OCR tensor methods MCMC


📫 Contact

denissamatov470@gmail.com · Telegram @SamatovDS · Tomsk, Russia — open to remote work and relocation.

Pinned Loading

  1. tensor-based-modal-decomposition-method tensor-based-modal-decomposition-method Public

    Tucker/HOSVD-based tensor modal decomposition and QR sparse-sensor placement for reservoir field reconstruction (arXiv:2607.09687)

    Python 1

  2. yandex-workspace-mcp yandex-workspace-mcp Public

    MCP server for Yandex Disk & Wiki — security-first, read-only by default, audit-logged

    Python 1

  3. llm-adviser-case-study llm-adviser-case-study Public

    Sanitized architecture case study: agentic engineering knowledge platform — deterministic document processing, traceable RAG, hybrid retrieval

  4. mcp-capguard mcp-capguard Public

    Assert that an MCP server's configuration profile exposes exactly the tools it should, and none it shouldn't — permission/capability boundary testing, not schema drift.

    Python

  5. agentic-architectures agentic-architectures Public

    23 agentic architecture patterns (RAG, GraphRAG, multi-agent) implemented with LangChain/LangGraph

    Jupyter Notebook 1

  6. weaviate/weaviate-python-client weaviate/weaviate-python-client Public

    A python native client for easy interaction with a Weaviate instance.

    Python 228 155