Research

Research is most useful when its assumptions and mechanisms are clear. This section connects model and systems behaviour to the decisions engineers make about performance, reliability, and infrastructure.

How to read an analysis

Separate the original claim from the evidence, identify the workload and assumptions, and ask what changes in a different environment. Engineering implications should follow from those limits.

Selected systems deep dives

• Before the Pod Exists: How Kueue Decides Whether an AI Job May Run
• The KV Cache Is Your Real LLM Serving Budget
• When an Agent Calls a Tool, What Happens to Its KV Cache?
• Evaluate the Agent Trajectory, Not Just Its Final Answer

These existing articles provide starting points. New research articles will appear below as they are classified and published. For guided background, choose a path under Learn.