At Visa Inc., I developed an end-to-end credit card fraud detection pipeline, achieving 97.13% ROC-AUC and 76.63% PR-AUC on highly imbalanced transaction data. I also automated threshold optimization to support business-driven performance trade-offs.
I developed an MLOps monitoring framework with PSI-based data drift detection and severity-based alerts. I implemented chunked batch inference with experiment tracking and pytest-based testing.
At Mercedes Benz Research and Development, I developed deep learning models for 3D obstacle detection and panoptic segmentation using LiDAR point-cloud data. I also developed RAG-based LLM chatbots for enterprise document retrieval and supported NLP-based quality-control workflows that saved 300+ hours of manual effort per week.
