I'm building practical AI systems for unstructured data, from Romanian legal question answering to document-layout parsing and autonomous-driving perception.
For a 32,920-document legal corpus, I designed a RAG pipeline with BM25 retrieval, metadata-aware reranking, noise filtering, domain-adaptive pretraining, and teacher-student distillation. The system reached 49.02% RAG accuracy on RoGemma-7B.
I've also engineered OCR pipelines for Ukrainian scene text and a monocular obstacle-detection pipeline that achieved a 0.764 overall F1-score and 0.784 average IoU on a selected BDD100K validation subset. My work spans Vision Transformers, Faster R-CNN, MiDaS, temporal tracking, and evaluation workflows.
I'm completing a Master's degree in Artificial Intelligence at Politehnica University of Bucharest and previously worked as a Robotic Cell Engineer at Ada Computers SRL. I bring Python, cloud AI tooling, model evaluation, and engineering discipline to machine-learning problems.
