A profile built with rigour, not a language trick.
Behind every conversation lies precise collection and verification work: we don't generate a plausible character, we build the most complete profile possible.
Collection
Texts, voice recordings, answers to targeted questions: every source the person (or whoever represents them) chooses to share becomes part of the profile.
Psycholinguistic profiling
We analyse vocabulary, sentence structure, recurring topics and moral positions to identify the traits that make a way of expressing oneself recognisable. Not just the words used.
Completeness check
Every profile is measured against how much material is needed to be reliable: when too much is missing, we say so instead of filling the gaps at random.
Model construction
The verified profile trains or guides a dedicated language model, never shared between different people.
Consent-based activation
The model stays inactive until whoever has the right (the person themselves, or whoever represents them under our consent rules) authorises its use.
This process is grounded in an internal research framework that measures how genuinely complete a profile is before it can be considered reliable. The technical details remain confidential to our researchers' work, but the principle is public: we never publish an echo we know to be poorer than the real person.
