Rahul Sathiyababu

• Worked on Continous Development of ML models for human activity recognition and pose estimation. • Optimized ML data ingestion pipeline, reducing upload time by 60% that improved performance. • Used CVAT application for bounding box, label and key point annotation to create the custom dataset. • Utilized ClearML to managed, version datasets and track experiments to ensuring data consistency and facilitating reproducibility.

Worked in 3D keypoint-based human action recognition using supervised and semi-supervised learning methodologies with curriculum learning and pseudo-labeling techniques for Transformer models. Worked in Self-Supervised training for Gated-Transformer network using masked vision model technique for 3D reconstruction.

Developed and deployed a machine learning-based dashboard, improved the quality of in-car software by 10%. • Analyzed and transformed large-scale car telemetry data using Vaex, Pandas, to extract meaningful insights. • Preprocessed and consolidated data from multiple servers, optimizing the execution time by 20%. • Implemented ML data pipeline for predicting software cash using ML models with preprocessing and analyzing.

Designed and implemented multisite web application with self-care features that allowed customers to access and analyse their data, providing insights, facilitating informed decision-making improved performance by 50%. • Created user-friendly data visualization components, enhancing data exploration and saving 30% of time.

Thesis Topic: "Enhanced Spatial-Temporal Transformer for Human Hand Action Recognition in Assembly Line" Designed and compared three novel Transformer models that combine spatial and temporal features: Serial Transformer, Parallel Transformer with self-attention and Cross Transformer with cross-attention mechanism.