ND-LLM
A neurodevelopment-specific language model.
A model trained on longitudinal neurodevelopmental records rather than the open internet, built to make a decade of fragmented history legible in the minutes before an appointment.
Domain
Pediatric neurodevelopment
Trained on
Longitudinal care records
Role
Assistive, not diagnostic
Stage
In development
01Why a domain-specific model
General-purpose models learn from the distribution of text on the internet. Pediatric neurodevelopment is thinly represented there, and the part that is represented skews toward the dramatic and the atypical rather than the ordinary range of development.
The ND-LLM is built on a different and much narrower distribution: longitudinal records of how children actually develop over years, and how they respond to intervention.
02What we are trying to make it good at
Preparing a case before a consultation, so a clinician spends their time on judgement rather than on reconstructing a history from scattered documents.
Surfacing changes in a developmental trajectory that are easy to miss between visits spaced months apart.
Summarising a long, fragmented record without flattening the detail a specialist needs.
03What it is explicitly not
It is not a diagnostic device and we do not present it as one. Diagnosis is a clinician's call, made with the child in front of them.
It is also not a replacement for the dataset underneath it. The value compounds in the longitudinal record itself; the model is a way of reading that record.
See what Deep Connection is really building.
Whichever side of this you're on, there's a way in.