VetScribe AI

VetScribe AI

Ambient AI Scribe for Veterinary Medicine Turn exam conversations into structured SOAP notes — hands-free, in real time.

The Problem

Veterinarians spend 2–3 hours per day on clinical documentation. Manual typing during exams is distracting, error-prone, and takes time away from patient care. Existing voice-to-text tools lack clinical context, species awareness, and integration with practice management systems.

The Solution

VetScribe AI is an ambient AI scribe purpose-built for veterinary medicine. It listens to exam conversations, identifies speakers (vet vs. client), and automatically generates structured SOAP (Subjective, Objective, Assessment, Plan) notes — species-aware, citation-grounded, and ready for the veterinarian's review in under 90 seconds.

Key Capabilities

CapabilityDescriptionLive Audio CaptureStart, pause, resume, and end capture sessions. Browser microphone or exam room mic support.Speaker DiarizationIdentifies who is speaking (veterinarian, client, vet tech) with ≥95% accuracy at SNR ≥15dB.SOAP Note GenerationAI-powered, species-aware SOAP notes with grounded citations. Sections stream incrementally as generated.SNOMED-CT CodingAuto-suggests veterinary clinical codes and invoice line items from finalized notes.Template ManagementCreate visit-type-specific templates (wellness, sick, dental) that structure SOAP output.PIMS IntegrationPush finalized notes, codes, and billing to ezyVet and Cornerstone with retry and backoff.Client SummariesGenerate plain-language summaries at ≤6th grade reading level — delivered via email, SMS, or print.Voice CommandsHands-free corrections: flag for follow-up, change dosage, add note, schedule reminder.Multi-TenantRow-level security isolates data per practice. Practice-manager role for cross-location oversight.HIPAA-AlignedPHI redaction (18 HIPAA identifiers), encryption at rest/in transit, append-only audit trail, BAA-gated model tuning.

Architecture

┌─────────────┐ REST/SSE ┌──────────────┐ gRPC ┌─────────────────┐ │ React SPA │ ◄──────────────► │ .NET 10 API │ ◄───────────► │ Python AI │ │ (Vite/PWA) │ │ Clean Arch │ │ (FastAPI) │ │ │ │ CQRS/MediatR│ │ faster-whisper │ │ Auth: OIDC │ │ EF Core │ │ LLM (OpenAI) │ │ Cache: RQ │ │ Serilog │ │ PII Guardrail │ └─────────────┘ └──────┬───────┘ └─────────────────┘ │ ┌─────────┴─────────┐ │ │ ┌────────▼──────┐ ┌────────▼──────┐ │ PostgreSQL │ │ Redis │ │ + pgvector │ │ (Cache) │ └───────────────┘ └───────────────┘

Tech Stack

LayerTechnologyFrontendReact 19, TypeScript, Vite, Tailwind CSS 4, TanStack Query, ZustandBackend.NET 10 / C# 14, ASP.NET Core, MediatR, FluentValidation, EF CoreAI ServicePython 3.12, FastAPI, faster-whisper, OpenAI, AnthropicDatabasePostgreSQL 16 + pgvectorCacheRedis 7MessagingRabbitMQ / Azure Service Bus via MassTransitInfrastructureDocker, Kubernetes (AKS), Terraform, HelmObservabilityOpenTelemetry, Serilog, Azure Monitor, GrafanaAuthenticationOpenID Connect (Entra ID / Auth0)

Performance Goals

MetricTargetAPI readsp95 < 200msAPI writesp95 < 400msSOAP generation first-token< 1.5s (streamed)Full SOAP note< 6sSpeaker diarization accuracy≥95% at SNR ≥15dBSOAP note acceptance rate≥90% without significant reworkSystem availability99.9%

Security & Compliance

• HIPAA-aligned: Encryption at rest (AES-256) and in transit (TLS 1.3)

• PHI Redaction: All 18 HIPAA identifiers replaced with category labels before AI processing

• Audit Trail: Append-only logging of all security-relevant actions

• Multi-Tenant Isolation: Row-level security with tenant key on every database row

• BAA Gating: Model tuning from correction data requires signed BAA opt-in per practice

• SBOM: Software bill of materials generated per build

Deployment

• Cloud-native on Azure Kubernetes Service (AKS)

• Docker containers with Helm charts

• Infrastructure-as-Code via Terraform

• CI/CD via GitHub Actions with automated testing, SAST, and container signing

VetScribe AI — Reducing documentation burden, one exam at a time.