18.10.2026 aktualisiert
100 % verfügbar
Data Scientist & Machine Learning Engineer
Budva, Montenegro
Nur Remote
Information Security, Institute of Cryptography, Telecommunications & Informatics,Moscow (1997-2002)Skills
Sql Data WarehousePython
Machine Learning & AI
Expertise in LLMs and Transformers (LLaMA 2, HuggingFace), Deep Learning with PyTorch and Keras, Computer Vision using YOLOv11 and OpenCV, Reinforcement Learning with actor-critic and experience replay, as well as Scikit-learn and activation engineering techniques.
Data Engineering & SQL
Proficiency in SQL databases including MS SQL, Oracle, PostgreSQL and MySQL, with exposure to BigQuery and Snowflake. Python data stack covering Pandas, NumPy, Matplotlib and Seaborn, plus statistical analysis and data modeling (3NF, star/snowflake, OLAP).
ETL / BI / DWH Development
Hands-on experience with Pentaho Data Integration, Pentaho Business Analytics, SAP BusinessObjects and GoodData. Design and implementation of DWH architectures, data migration, reconciliation and data-quality frameworks at scale (1T+ records).
Python Engineering & APIs
Development of Flask REST APIs, Telegram Bot integrations, memory-efficient inference pipelines, modular and plugin architectures, and Git-based workflows on Linux environments.
Computer Vision Pipelines
End-to-end object-detection pipelines for high-resolution aerial imagery using YOLOv11, tiling strategies, prediction re-assembly, non-max suppression, and deployment via Flask REST API with Telegram bot integration.
LLM Internals & Activation Engineering
Training-free control of LLaMA 2 outputs via steering vectors injected into attention heads, automated search over layer-head pairs, cosine-similarity gating, and PyTorch forward hooks with VRAM-aware memory safety.
Reinforcement Learning & Trading Systems
Design of RL layers using actor-critic, experience replay and headless worker on a shared MDP abstraction. Signal engines combining technical indicators (ADX, ATR, SMA, RSI) with volatility-based filtering and ML candidate-rankers with offline back-testing.
Weights & Biases / MLOps
Instrumented training and evaluation using Weights & Biases, including loss tracking, mAP, confusion matrices and sample-level visualizations for model monitoring and experiment management.
Domain Expertise
Applied experience across investment banking, insurance, mortgage lending, HCM/HR analytics, crypto and trading signals, and search and rescue computer vision applications.
Expertise in LLMs and Transformers (LLaMA 2, HuggingFace), Deep Learning with PyTorch and Keras, Computer Vision using YOLOv11 and OpenCV, Reinforcement Learning with actor-critic and experience replay, as well as Scikit-learn and activation engineering techniques.
Data Engineering & SQL
Proficiency in SQL databases including MS SQL, Oracle, PostgreSQL and MySQL, with exposure to BigQuery and Snowflake. Python data stack covering Pandas, NumPy, Matplotlib and Seaborn, plus statistical analysis and data modeling (3NF, star/snowflake, OLAP).
ETL / BI / DWH Development
Hands-on experience with Pentaho Data Integration, Pentaho Business Analytics, SAP BusinessObjects and GoodData. Design and implementation of DWH architectures, data migration, reconciliation and data-quality frameworks at scale (1T+ records).
Python Engineering & APIs
Development of Flask REST APIs, Telegram Bot integrations, memory-efficient inference pipelines, modular and plugin architectures, and Git-based workflows on Linux environments.
Computer Vision Pipelines
End-to-end object-detection pipelines for high-resolution aerial imagery using YOLOv11, tiling strategies, prediction re-assembly, non-max suppression, and deployment via Flask REST API with Telegram bot integration.
LLM Internals & Activation Engineering
Training-free control of LLaMA 2 outputs via steering vectors injected into attention heads, automated search over layer-head pairs, cosine-similarity gating, and PyTorch forward hooks with VRAM-aware memory safety.
Reinforcement Learning & Trading Systems
Design of RL layers using actor-critic, experience replay and headless worker on a shared MDP abstraction. Signal engines combining technical indicators (ADX, ATR, SMA, RSI) with volatility-based filtering and ML candidate-rankers with offline back-testing.
Weights & Biases / MLOps
Instrumented training and evaluation using Weights & Biases, including loss tracking, mAP, confusion matrices and sample-level visualizations for model monitoring and experiment management.
Domain Expertise
Applied experience across investment banking, insurance, mortgage lending, HCM/HR analytics, crypto and trading signals, and search and rescue computer vision applications.
Erfahrung
Berufserfahrung0 - 5 Jahre
Freelancer seit2023
Sprachen
GermanGrundkenntnisseEnglishverhandlungssicherFrenchGrundkenntnisseItalianGrundkenntnissePortugueseGrundkenntnisseRussianMutterspracheSpanishgut
Projekthistorie
Self-directed applied R&D across LLM internals, computer vision and decision-automation systems. Projects include LLM steering and fact editing via activation engineering on LLaMA 2, a YOLOv11 computer-vision pipeline for search and rescue (LizaAlert Detector), a crypto signal bot with ML ranking and RL-ready trading architecture, and a universal converter modular API-first toolkit. All projects open-sourced on GitHub.
Owned ETL development and reporting across 360+ client requests, reducing turn-around from months to weeks. Restructured the product backlog, reducing average open tasks from 100+ to 50-70. Delivered self-service analytical dashboards in GoodData with time-period and segment drill-down. Stack: MySQL, PostgreSQL, Pentaho Data Integration, Pentaho BA, GoodData.
Led datamart and ETL improvements on the corporate-finance transaction showcase; achieved a ~20% productivity gain and materially reduced erroneous records feeding P&L reporting. Stack: SAP BusinessObjects, Oracle, SQL.