Niranjan Madhu

Co-founded a venture focused on sensor fusion and reinforcement learning. Analyzed data from multiple sources including images and edge sensors, and developed RL policy training modules in JAX for faster inference.

Designed and deployed CatBoost and XGBoost models for insurance premium prediction, achieving MAE of ~$350 and MAPE of 23-25% for Personal Auto and MAE of ~$200 with MAPE of 12-15% for Homeowner. Led migration to an internal ML-driven quoting engine, delivered CTR uplift of 14-18%, and built an LLM-based semantic mapping system for NAIC codes.

Improved RoBERTa-based classification F1 score by 219% through error analysis and fine-tuning. Built a Concept Drift Detection module using n-gram distributions and RKHS semantic distance, and developed a CNN + YOLO-based hierarchical document parser.

Implemented LLM research workflows including RAG, IR, and contextual augmentation into production systems. Optimized GPU-to-CPU inference architectures for cost-efficiency and built fraud detection systems achieving >0.90 recall with XAI reports.

Helped clients with new features to reduce fraud and created a fraud detection model with recall greater than 0.90. Documented feature causality using post hoc explainable AI methodology.

Grade: 3.4

Pursued a triple major in Computer Science, Mathematics, and Statistics, achieving a GPA of 3.4 and ranking in the top 10% of the class.

Completed the Neural Network and Deep Learning certification by Andrew Ng.

Completed the DAT273x: Data Science Research Methods: Python Edition certification.

Completed the PH525.1x: Statistics and R certification.

Completed the Improving Deep Neural Networks: Hyperparameter tuning, regularization, and optimization certification by Andrew Ng.

Completed the Convolutional Neural Networks certification by Andrew Ng.

Completed the Natural Language processing in TensorFlow certification by Andrew Ng.