Built for a media monitoring use case the goal was to automatically tag thousands of daily news articles by topic (politics, tech, finance…) and flag sentiment trends over time. I fine-tuned a BERT-based model (specifically bert-base-uncased) on a labeled news corpus using HuggingFace Transformers. Classification hit 94% accuracy across 8 categories. On top of that, I layered a sentiment module using VADER for speed on high-volume data. The final output was a dashboard where you could track how sentiment around any topic shifted week over week. Very useful for PR teams or investors.