Sunil

Built a multi-class NLP classifier that tags incoming support tickets by intent and urgency, then routes them to the right queue. Trained on 60,000 historical tickets with a fine-tuned DistilBERT model served through FastAPI, reaching 91% macro F1 and auto-routing 42% of volume. Median first-response time dropped from 6.5 to 4.7 hours.

Aspect-based sentiment analysis and topic modeling over 400,000 product reviews to surface recurring complaints per category. Combined BERTopic for theme discovery with a fine-tuned sentiment model, then delivered a Streamlit dashboard merchants use weekly. Flagged three defect clusters that drove a packaging change and a measurable drop in return rate.