Horus: An AI Copilot That Shows Its Work
TropX pairs a deterministic clinical engine with a language copilot. Every AI answer is labeled with its evidence base and a reminder to confirm clinically.
An AI feature in a clinical tool has to be checkable. In TropX, the analysis itself comes from a deterministic engine that runs on measured kinematics. The assistant, Horus, reads that analysis and explains it in plain language, with the underlying numbers in view.
Ask why the left leg is slower, and the answer refers to the same measurements the session report shows. Every conversation is headed by a disclosure line stating how many recordings the analysis is based on, followed by a reminder to confirm clinically.
Memory that surfaces as evidence
Horus keeps a semantic memory of past sessions. When a question touches history, for example whether a patient has plateaued before, the assistant recalls the closest matching sessions and embeds them in the chat as session cards showing the left/right waveform, the OPI, the grade, and the date.
The clinician can open any recalled session and replay it. Because the referenced session arrives inside the answer, a claim about the past can be checked on the spot.
The division of labor
The split is deliberate. Language models are good at reading and explaining. Computing a joint angle is a job for the measurement pipeline, and TropX never hands that job to a language model.
- The clinical engine is deterministic. The same session produces the same metrics, scores, and flags every time
- The copilot explains and compares in natural language, in the clinician's terms
- Cross-session analysis surfaces recurring patterns such as persistent asymmetry or a declining metric
- Every AI surface carries the confirm-clinically label, and the clinician has the final word
The result behaves like a careful colleague. It remembers the full history, it answers with its sources attached, and the clinical decision stays with the clinician.