Many "emotional" systems try to role-play. Velin is built around signals instead: mood, risk, pace, and attachment tendencies.
The model does not love or hate anyone. It simply estimates how safe and stable a conversation is and gives products a small set of dials to react appropriately. For example, it can slow down, suggest a break, or avoid specific kinds of replies when risk is high.
This is more boring than role-play, but we care more about not harming vulnerable users than about being entertaining.