Маріанна

MARIANNA SYCHAK
[открыть контакты](см. выше в блоке «контактная информация») | [открыть контакты](см. выше в блоке «контактная информация»)
SUMMARY
Motivated 3rd-year Cybersecurity student with a strong technical background and analytical mindset. Experienced in data processing, quality assurance, and technical documentation through university projects. Proficient in Office 365, Linux/Windows administration, and Python. Possess deep mathematical knowledge and a keen eye for detail, along with growing expertise in Machine Learning algorithms for data analysis and anomaly detection.
EDUCATION
Ivan Franko National University of Lviv, Lviv Bachelor's Degree | Sep. 2023 - Jun. 2027
SKILLS
• Programming Languages: Python, C++, HTML (basic).
• Machine Learning & Deep Learning: Classical ML (classification, regression, clustering via Scikit-learn), Deep Learning (PyTorch), NLP & Speech-to-Text (Whisper models), Data Analysis & Feature Engineering (Pandas, NumPy), with practical experience in Fraud Detection and Automated Content Moderation.
• Languages: Ukrainian (Native), English (B2), Polish (A2).
PROJECTS
Automated Content Moderation & Audio Processing Pipeline
• Designed and developed an automated text and audio content filtering system by building an end-to-end Machine Learning pipeline in Python.
• Integrated and optimized Whisper models for high-accuracy Speech-to-Text (STT) processing, enabling seamless voice-to-text transformation and analysis.
• Implemented NLP techniques for feature extraction, tokenization, and text vectorization to identify and flag restricted content.
• Engineered robust data processing pipelines using Pandas and NumPy to handle multimodal datasets, optimize feature selection, and evaluate model performance.
• Applied classical ML algorithms alongside deep learning approaches to balance classification speed and evaluation accuracy in a production-like environment.
Skelarxmono Hackathon | Hackathon Team Project
• Designed and engineered a user-level anti-fraud system to detect fraudulent accounts by shifts in transactional behavior rather than single operations, formulating it as a binary classification task.
• Developed comprehensive behavioral profiles by aggregating transactional histories into high-impact features capturing activity intensity, risk distribution, execution velocity, and failed transaction patterns.
• Engineered advanced contextual and risk features, including initial latency from registration, time-of-day risks, and geospatial fraud probabilities.
• Addressed severe class imbalance through iterative feature engineering and architectural benchmarking, optimizing the F1-score to strike a precise balance between precision and recall.
• Trained, optimized, and deployed a LightGBM gradient boosting model to capture non-linear feature interactions, successfully outperforming baseline Logistic Regression and Random Forest architectures.
• Designed a production integration strategy that pairs the ML model with rule-based mechanisms to score user risks, enabling multi-tiered automated actions like temporary restrictions, manual review triggers, or step-up authentication.