Developed LLM-based systems using RAG to intelligently integrate data for problem-solving (e.g., medical diagnosis).
Built custom AI solutions including vision-based applications (gesture recognition, object tracking), fraud detection (from chess to MMOs), and 3D data processing (laser scanning, scene reconstruction).
Worked on the 'rift.guru' project, training backend AI to learn strategic character combinations in League of Legends and provide in-game advice.
Trained neural networks for self-driving in virtual environments, tackling challenges like environmental complexity, real-time control, and limited training data.
Explored combining Mixture-of-Experts (MoE) with single-network multi-configuration models to enhance LLM specialization while preserving shared general knowledge.