Longitudinal analysis

Any decent clinical AI today is capable of looking at one session and producing a reasonable synthesis of what happened in those 60 minutes. Longitudinal analysis is something else: it is the reading that crosses session 14 with session 8 and with session 22 to identify that the same theme returns three months later, transformed but recognisable. It is the gaze that detects that a therapeutic alliance began to build around session 5 and has stayed stable ever since. It is the recognition that an intervention made in February produced a change that has only become visible now, in May. That reading is mathematically expensive. Looking at one session is ~30K tokens of context; looking at a 30-session process is ~900K tokens. Brute-forcing it every time is economically unviable. The way nexmin solves this is the client's wiki: a compact, structured representation of the full process that the system keeps up to date and that Pensa reads as context before looking at the new session. Result: the engine has real longitudinal memory without reprocessing the whole history every time. What this changes for the therapist: they stop being the only one mentally holding the trajectories of twelve or fifteen active clients. When they have not seen someone for three months and that person comes back, they do not start from zero — the system presents them with the reading of the process, with the nuances they themselves articulated when they were sharper. That return is one of the clearest practical differences between nexmin and a transcription + summary tool.

Related terms

• Pensa
• Clinical cartography
• Phenomenological identity