The Problem With Veterinary Telemedicine at Scale
Veterinary telemedicine sounds straightforward until you start counting the moving parts. You have two distinct user types — pet owners and veterinary experts — with completely different workflows, different interfaces, and different expectations from the same underlying system. A pet owner wants a frictionless experience; a vet expert needs a clinical tool. The same platform has to serve both without compromising either.
Add real-time video and voice to that, and the surface area for failure grows quickly. A dropped call at a critical moment isn't a UX inconvenience — it breaks trust in the product. When PawSquad (part of IVC Evidensia) started scaling toward hundreds of thousands of users, the infrastructure had to be reconsidered from the ground up.
Our starting point was a clean architectural separation between the consumer-facing layer and the clinical layer, unified by a single backend API and real-time communication infrastructure.
Platform Architecture: What We Built
The platform spans three distinct surfaces, each engineered separately but communicating through a shared WebAPI layer built on .NET 10.
Multi-tenant · Multiple environments (dev / QA / UAT / production) · Any major database engine
The Real-Time Decisions
The choice to use Twilio Router alongside Twilio Video was deliberate. Routing a veterinary consultation is not the same as placing a VoIP call — you need to match pet owners to available experts, handle queue state, support fallback scenarios when a vet is unavailable, and manage session metadata throughout. Twilio Router gives you programmable call flow; Twilio Video gives you the media transport. Together they cover the full consultation lifecycle.
For everything that doesn't require media — async chat, queue notifications, status updates, document sharing — we use SignalR over WebSockets. This separation matters: media and signaling have different reliability profiles and different failure modes. Mixing them into a single transport would complicate both.
The result is a system where real-world call failure rates stayed below 3% at peak load. On the infrastructure side, our internal numbers came in under 1%.
AI in the Consultation Loop
Post-consultation reporting was a bottleneck. After every session, vets were expected to produce a clinical summary — transcript review, anamnesis, diagnosis notes, follow-up actions. Done manually, this was time-consuming and inconsistent between practitioners.
We integrated an AI processing layer that activates automatically at the end of every consultation, regardless of type — video, voice, or text chat. The output the vet receives includes a full transcript, a structured summary, a pre-built anamnesis section, and (for applicable consultation types) a structured starting point for diagnosis and recommended actions.
The vet reviews, adjusts, and confirms. The AI eliminates the blank page and the manual reconstruction — it does not replace clinical judgment.
Data protection was a non-negotiable design constraint. Before any content reaches an external AI service, it is fully anonymized: names, contact details, and any identifiers that could link content to a specific individual are stripped. What the AI processes is clinical content only. The anonymization layer is automatic, configurable per deployment, and leaves an audit trail. Under GDPR, this distinction — between identifiable and de-identified clinical content — is the difference between compliant and non-compliant AI integration in healthcare.
Platform as Product
PawSquad is not running a collection of features bolted together — it is a platform with a clear architectural spine: one API, one real-time layer, multiple surfaces. The decisions we made early — separate consumer and clinical interfaces, Twilio for media with SignalR for signaling, AI with anonymization rather than AI with raw data — held up under load and under regulatory scrutiny.
350,000+ pet owners and 120+ veterinary experts now use a system that didn't exist five years ago. Building it required solving problems in real-time communications, multi-tenant architecture, clinical workflow design, and AI integration — often simultaneously. That combination is where Smartnet's expertise sits.