Soham Dutta is a Principal Software Engineer at Red Hat and a full-stack AI engineer, building across the whole pipeline: data preparation, fine-tuning, retrieval, agent systems, and the inference and Kubernetes infrastructure that carries them to production. His current work centers on LLM serving performance, including KV cache reuse, memory tiering, and multi-replica scaling with vLLM, LMCache, and Valkey, plus the tenant isolation questions that shared caches raise.