At COMPELLIO S.A., I developed an LLM-powered question-answering system for PDF and JSON documents using transformer models including BERT, DistilBERT, and T5. I combined neural models with TF-IDF retrieval and rule-based logic to handle structured content, and fine-tuned models on domain-specific data.
My projects also include an award-winning self-driving car thesis, where I implemented Q-Learning obstacle avoidance and route tracking for simulated and physical systems. I’ve worked across data analysis, web and blockchain applications, and embedded systems, using tools such as Python, SQL, and C#.

