aditmodi.hashnode.devThe Unified GPU Platform: Running Slurm, Ray, and Kubernetes Inference on a Single EKS Cluster Without Scheduling ChaosThere's a meeting that happens at every organization the moment their AI ambitions outgrow their GPU budget. It usually involves three teams talking past each other. The HPC team says: "We need 32 GPU4d ago·32 min read
cloudtech.hashnode.dev👋 Everything about Cloud & Tech Newsletter "#58" ☁️❤👨💻Dear Cloud and Tech enthusiasts, Welcome to Everything about Cloud & Tech #58. There’s a question underneath everything in this edition that nobody says out loud but everyone is quietly asking: how do6d ago·18 min read
aditmodi.hashnode.devInference Engineering — Book ReviewSome books explain AI infrastructure with clean diagrams and tidy abstractions. This one pulls you into the engine room and shows you what actually happens between a prompt and a response — the memoryMar 3·6 min read
cloudtech.hashnode.dev👋 Everything about Cloud & Tech Newsletter "#57" ☁️❤👨💻Dear Cloud and Tech enthusiasts, Welcome to Everything about Cloud & Tech #57. There’s a pattern showing up across everything in this edition that I didn’t fully notice until I started writing the intFeb 23·16 min read
aditmodi.hashnode.devEKS Multi-Account Administration: From GitHub Issue to Production ArchitectureIn April 2022, a deceptively simple GitHub issue landed in the AWS containers roadmap: "Looking to start a discussion and get some feedback to help guide our strategy around helping customers administFeb 21·22 min read