Calmmas Apsys26

CALM-MAS 🧘🏽‍♂️🧘🏽‍♂️ has been accepted at ACM APSys 2026. We address the question: “What if we treat test-time compute as an elastic knob of agentic LLM applications, and regulate the amount of per-session computation depending on the congestion state of the serving engine?” Mouheb will present this idea and share our findings in Bangkok 🧘🏽‍♂️