Projects
Charactly
A multi-tenant platform that extracts a character's speaking style and lets you talk to them.
Problem
Talking to a character “in their voice” does not happen by telling the model their name. Speaking style — rhythm, word choice, what they leave unsaid — has to be extracted from something real.
Solution
Two ways in:
- From video: a YouTube channel or video link → audio download → local Whisper transcription → speaker separation → a separate character profile per voice.
- Screenwriter mode: a written script or character sheet → direct LLM analysis → profile plus casting suggestions.
The platform is multi-tenant: every tenant’s data stays inside its own boundary.