An imperative language for AI workload orchestration

A fault-tolerant LLM routing system that decouples inference from AWS Bedrock by routing prefill and decode tasks through SQS and ensuring zero-downtime scaling with a graceful drain sidecar.