Veterinary Surgical Documentation System logoVetucate AI

Pipeline / Technical

Eight-stage procedural pipeline.

Recording flows through fixed, ordered stages. Each stage has defined inputs, bounded outputs, and explicit constraints. No stage performs open-ended inference.

Stages

Recording through export.

Capture
01

Recording

Local capture of surgical video and audio. Write-only. No processing occurs during this stage.

02

Interval segmentation

Recording divided into fixed 30-second intervals. Each interval is independently bounded. Failures do not propagate across them.

Extract
03

ASR processing

Spoken procedural context extracted per interval: transcripts, numeric calls, timestamps. Literal transcription only. Missing audio records as null.

04

VLM extraction

Visual evidence extracted per interval: instruments, vitals, patient positioning, environmental markers. Observation only, no inference.

Synthesis
05

Evidence synthesis

ASR and VLM outputs merged into a validated per-interval evidence block. Fields not reliably extracted record as not recorded. Never fabricated.

06

Timeline generation

Navigable chronological record assembled from the sequence of evidence blocks. Generated from structured evidence only.

Output
07

Report generation

Structured post-operative report compiled from extracted evidence. Gaps are surfaced explicitly, not filled.

08

Export packaging

All artifacts bundled into a self-contained package: timeline HTML, synchronized recording, report, manifest. Operator-approved transfer only.

Vetucate does not autonomously infer surgical actions. Each stage extracts from captured evidence only.

Evidence blocks

Immutable units.

Each 30-second interval produces a sealed evidence block, independently verifiable and independently recoverable.

Definition

An evidence block = chunked capture, aligned transcripts, model outputs, and a cryptographic hash.

Immutability

Write-once. Updates create new superseding blocks linked via an append-only audit trail.

Timestamps

Every field stores start/end times for forensic fidelity.