{ "intent_classifier_ms": 0.476, "anomaly_detector_ms": 10.104, "kb_retrieval_ms": 0.735, "note": "LLM generation latency depends on the external Inference API call and is measured live in the app, not benchmarked here." }