Worked on the pytorch/torchrec repository to deliver distributed inference sharding support for FeatureProcessedEmbeddingBagCollection (FPE_EBC) models on High Bandwidth Memory (HBM). Developed logic to ensure sharding was restricted to HBM, avoiding cross-CPU sharding to maintain memory locality and optimize performance. Implemented environment-aware distributed inference sharding, including propagation of tensor broadcasting events to support scalable inference deployments across diverse hardware setups. Utilized Python, PyTorch, and distributed systems expertise to enable hardware-aware inference workflows. The work focused on building foundational infrastructure for scalable, efficient inference, addressing both technical constraints and business value in large-scale machine learning environments.