Research / Technical
Evidence-bound documentation research.
Vetucate combines local capture, evidence-linked clinical entries, professional wording checks and clinician-approved outputs.
Separate evidence.
One case record.
What this system investigates
A source sentence for each spoken fact. A fixed clinical record structure. Checks against the captured words before finalization. A narrative traced sentence by sentence to record entries.
Procedure context guides extraction while remaining separate from patient evidence. The owner’s patient-specific summary is traced to the approved record.
Explicit boundaries
Vetucate does not diagnose, predict outcomes or infer an unstated procedure. Clinical authority stays with the veterinarian.
The procedure library cannot supply patient facts. Silence is not converted into a negative finding. Missing information and conflicting values remain reviewable.
Testing scope
Hundreds of hours of replay testing across numerous cases from the veterinary surgery database, plus live recordings on the appliance itself.
The pipeline is assessed against a hand-built account of spoken facts. Records are compared case by case before changes ship. 184 automated regression tests run before every release.
Reported performance and recovery
Reported single-camera finalization: 69 seconds for a 31-minute case, 92 seconds for a 64-minute case and about two minutes for a two-hour case. Video assembly runs in parallel.
Recovery was demonstrated after a live-case interruption, with repeatable records and ledger identifiers. Every interval stayed within its thirty-second processing bound in a 64-minute live case and a full replay of the two-hour case.
Current limits
Monitor extraction has been demonstrated on captured frames, not a live theatre monitor. Clinic-audio recall figures and updated dual-camera benchmarks are not published.
The draft can contain transcription errors, including drug names. Clinician review is required. Missing values cannot yet be typed into record fields on the appliance.