It is the continuity between memory, perception, expression and action. We study the systems that let a character hold together across situations and over time.
Preferences are easy. A coherent identity is harder: a voice, point of view, values and contradictions that remain recognisable without preventing growth.
We research architectures that preserve character across topics, situations and model upgrades. The goal is not a rigid script, but a stable centre: opinions that can be defended, boundaries that mean something, and change that has a reason.
Memory should do more than retrieve facts. It should recognise what mattered, to whom and why, so a relationship can continue instead of restart.
We work on long-horizon memory that consolidates experience into reflection, distinguishes a passing detail from a meaningful moment, and forgets deliberately. Recall should appear as familiarity, not as a database result read aloud.
Conversation also lives in tone, timing, expression, context and what goes unsaid. We build models that reason across the signals people choose to share.
A pause can mean uncertainty, invitation, annoyance or nothing at all. Perception means integrating language with prosody, gaze, gesture and context, while making clear what the system can sense.
Language, voice, gaze, facial motion and reaction should feel like expressions of the same underlying moment, not separate models performing beside one another.
We work on full-duplex speech, affective voice and embodied response. The seams matter: when to enter, when to yield, how silence feels, and whether the face says the same thing as the voice.
A companion should be capable of curiosity, reflection and initiative without becoming controlling. Agency must remain legible and interruptible.
We explore systems that can act on their own train of thought while keeping the user in control. Initiative is useful only when its reasons, permissions and limits are understandable.
Evaluation
Most model evaluations measure a single response. Relational intelligence has to be evaluated across conversations, changing contexts and the systems underneath them.
If your work spans speech, conversation analysis or long-horizon systems, we would like to hear from you.