Machine Learning Engineer (Active Learning Labeling – BERT Fine-tuning)
As part of a customer support ticket classification project, I fine-tuned a multilingual BERT transformer using annotated support tickets. The active learning pipeline operated to minimize manual labeling effort while improving model performance iteratively. My work included both automated and selective human review of text samples for appropriate model supervision. • Labeled and validated at least 5,000 text interactions for intent and sentiment. • Employed active learning strategies to sample ambiguous texts requiring annotation. • Used HuggingFace and Python as the primary software stack for labeling and training. • Validated and prepared annotated data for further model improvement cycles.