How VetScan's AI works

Last updated: April 27, 2026

1. Our approach to triage

VetScan does not diagnose. We help dog parents decide — when to wait, when to call a vet today, and when to go to an emergency clinic right now.

Every assessment maps to one of three urgency levels:

• Watch at home — likely benign, monitor for changes
• See a vet within 24-48 hours — non-emergency, but warrants professional review
• Emergency — go now — time-sensitive condition; visit ER or call poison control

This three-tier system was developed in collaboration with practicing veterinarians and reflects standard clinical triage frameworks (similar to those used in human ER intake).

2. Training data sources

Our AI is trained on a curated combination of:

• Peer-reviewed veterinary literature — including JAVMA, Veterinary Clinics of North America, and other accredited sources
• Published clinical guidelines from AVMA, ACVECC, ASPCA Poison Control, and the Merck Veterinary Manual
• Anonymized case patterns reviewed and validated by our veterinary advisory board
• Continuous review — a panel of licensed DVMs reviews and corrects model output on a quarterly basis

What we do NOT train on: We never use user-submitted photos, descriptions, or chat messages to train future models. User data stays user data.

3. What our AI considers

For each triage assessment, the model evaluates:

• Symptom characteristics — type, severity, duration, location, frequency
• Dog profile — breed, age, weight, vaccination status, known conditions, recent surgeries
• Time-sensitive flags — known toxin ingestion, breathing difficulty, severe trauma, suspected bloat, suspected heatstroke
• Behavioral changes compared to the dog's baseline reported by the owner
• Co-occurring symptoms — single symptoms vs. multi-symptom presentations are weighted differently

4. Confidence scores and escalation rules

Every recommendation comes with a confidence score (0-100%). Built-in safeguards include:

• If confidence is below 75%, the system automatically recommends consulting a veterinarian rather than offering a "watch at home" verdict.
• High-risk symptoms (e.g., labored breathing, collapse, suspected toxin ingestion, bloat signs, eye injury) always escalate to emergency, regardless of model output.
• Puppies under 16 weeks and seniors over 10 years are flagged for lower thresholds — symptoms that may be "watch at home" in a healthy adult become "see vet today" in vulnerable life stages.

5. What our AI does NOT do

We are deliberately narrow about scope:

• We do not diagnose conditions — only suggest urgency tiers
• We do not interpret X-rays, ultrasounds, blood work, or imaging
• We do not prescribe, recommend, or adjust medication dosages
• We do not handle veterinary medical records or replace your vet relationship
• We do not claim to be a licensed veterinarian or provide veterinary medical advice in the regulated sense

6. Continuous improvement

Veterinary medicine evolves. Our model evolves with it:

• Quarterly review — veterinary advisory board audits a sample of model outputs every quarter
• User feedback loop — owners can report when AI guidance didn't match what their vet later found; these reports feed back into training data
• Updated training data every 6 weeks to incorporate new clinical guidelines and reported edge cases
• Adversarial testing — we routinely run "red team" scenarios to find failure modes before users do

7. Transparency commitments

We commit to:

• Showing uncertainty — when the model isn't confident, we say so, plainly
• Plain-language disclaimers on every recommendation
• Open advisory board — see our veterinary advisors
• Research collaboration with academic veterinary schools (in progress)
• Public limitations log — known weaknesses published and updated

8. Questions

For questions about our methodology or to request more technical detail (we maintain a longer technical document for veterinary partners):

Email: hello@vetscan.app