Machine Learning Software Engineer at Mindtree

Built NLP models for customer query intent classification using the Hugging Face framework by transitioning from TF-IDF with Logistic Regression to a fine-tuned RoBERTa model that improved macro-F1 by approximately 25 percent and reduced manual triage of minority-class queries. Implemented hybrid fraud detection models combining cost-sensitive supervised algorithms with unsupervised anomaly detection techniques by benchmarking Logistic Regression and Random Forest baselines against XGBoost, Isolation Forest, and autoencoders to increase PR-AUC by nearly 28 percent and improve early fraud identification. Engineered a transformer-based BERT NER solution that used OCR-extracted text and passage tagging to identify clauses such as contract term and renewal term from PDF agreements. Collaborated with cross-functional teams to preprocess and structure raw contract data, and fine-tuned BERT using labeled inputs from the Legal team, improving model accuracy by 15% and reducing legal review tim