Worked on the pytorch/torchrec repository over a two-month period, focusing on enhancing training and evaluation pipelines using Python and machine learning techniques. Refactored the training pipeline to introduce common batch-processing methods, improving code reusability and maintainability while enabling seamless integration of latency tracking across multiple pipelines. Later, delivered enhancements to the hybrid training-evaluation workflow by implementing model-based evaluation detection and robust data-draining mechanisms, which ensured reliable processing of evaluation batches and prevented data loss. Extended the pipeline framework to support optimized sparse-distributed evaluation, contributing to improved scalability, reliability, and traceability in complex data processing and pipeline development scenarios.