Most 'AI understanding' stops at a label. KKMatch treats understanding as a continuous, evidence-linked process: observe signals, infer carefully, attach confidence, and let the person correct you.
Voice, language, and (with consent) video signals are normalized into behavioral features — never equated directly with emotions or states.
Every inference reports a confidence and an alternative explanation. Slow speech is not 'sadness'; looking away is not 'lying'.
You can see the evidence and challenge any conclusion. Understanding improves through your correction.
Q: Does AI read my emotions from my face?
A: No. Video is used only as an interaction signal (presence, attention, expressiveness), never as a diagnosis of feeling.
Q: Why show confidence?
A: Because behavioral inference is probabilistic. Confidence and alternative explanations prevent overclaiming.
Q: Who controls the model?
A: You do. Corrections and deletions are supported end to end.