Ekka: Automated Diagnosis of Silent Errors in LLM Inference

A University of Washington-affiliated project presents Ekka, a system for automatically diagnosing silent errors during large language model inference. The available record identifies the problem and project name but does not provide verified details about the diagnostic method, fault model, evaluation setup, supported inference stacks, or accuracy. This item is therefore best treated as an early research lead rather than evidence of a validated production solution.