AI Safety & LLM Security
Prompt Injection in Defended Systems
Evaluation framework for testing prompt injection attacks against LLMs with defensive mechanisms. Benchmarks attack success rates across multiple defense strategies.
Adversarial Attacks on LLM-as-a-Judge
Tools and experiments for evaluating the vulnerability of LLM-based evaluation systems to prompt injection and adversarial manipulation.
Trojan Detection in LLMs
Methods for identifying backdoors and trojans embedded in large language models. Solution code from the Trojan Detection Challenge.
NLP & Text Analysis
RuATD: Russian AI Text Detection
Machine learning pipeline for detecting AI-generated text in Russian. Built classifiers to distinguish human-written and machine-generated content.
Low-Resource Language Classification
Fine-tuning multilingual pretrained models (mBERT, XLM-RoBERTa) for sentiment analysis in African languages with limited training data.
Medical AI & Computer Vision
Glioma Grading from MRI
Deep learning model for non-invasive brain tumor grading from MRI scans. Classifies glioma grades to support clinical decision-making in neurosurgery.
MR-guided Glioma Typing
Machine learning pipeline for MRI-based non-invasive glioma classification, developed in collaboration with the Burdenko Neurosurgery Institute.
Neurosurgical Instrument Segmentation
Computer vision system for tracking and segmenting surgical instruments in video to assess microsurgical skill and support training.
Competition Solutions & Tools
Kaggle Competitions
Solutions for data science and ML competitions on Kaggle, covering computer vision, NLP, and tabular data challenges.
AI & ML Cheatsheet
Comprehensive reference guide covering mathematics, ML algorithms, neural networks, NLP, computer vision, and more. Used in teaching at Moscow State University.
Deep Learning Course Materials
Practical tutorials and exercises for deep learning, developed for courses at Moscow State University.