Deploying Disaggregated LLM Inference Workloads on Kubernetes

Build an open-source controller that handles disaggregated LLM inference (splitting prefill/decode phases) on standard Kubernetes clusters. Current tools usually treat inference as a monolith, wasting resources.

Suggested repo: DecoupleLLM

"Maximize your LLM throughput by decoupling prefill and decode stages."

Estimated effort: 180h