The Problem We Were Solving
Physiotherapists spend a disproportionate amount of their clinical time on work that is not clinical: documenting intake, searching for relevant protocols, writing session notes, and preparing referral letters. In a busy clinic, this administrative overhead can consume 30–40% of available appointment time — time that should be spent with patients.
The second problem is consistency. Clinical assessment quality varies with fatigue, time pressure, and the practitioner's familiarity with specific conditions. A less-experienced physiotherapist seeing a complex presentation may not immediately surface the most relevant evidence-based protocols — not because the evidence does not exist, but because accessing it in a time-pressured clinical environment is impractical.
Physiolaxy was built to solve both problems simultaneously: reduce administrative overhead and elevate clinical consistency — without replacing clinical judgement.
The Architecture: Agent-First from Day One
We made the decision early to build Physiolaxy as an agent-first system rather than a form-based application with AI features bolted on. This means the AI assessment engine is not a module — it is the primary user experience. The practitioner interacts with an intelligent intake system that asks the right questions in the right sequence, adapts based on responses, and builds a structured clinical picture in real time.
The core components:
- Intake agent — Guides the patient through a structured, adaptive symptom questionnaire. Question branching is driven by prior responses, not a fixed decision tree.
- Assessment engine — Takes the structured intake data and produces a differential assessment summary with confidence ratings and the reasoning behind each finding.
- Protocol recommendation engine — Cross-references the assessment against a curated database of physiotherapy protocols and clinical guidelines, ranking recommendations by relevance and evidence quality.
- Documentation agent — Generates session notes, referral letters, and compliance documentation from the clinical interaction — formatted to the practice's templates.
The Multi-Tenant Architecture
Physiolaxy needed to serve both solo practitioners and enterprise clinic networks from the same platform. The architectural requirement: complete data isolation between tenants, independent branding and configuration per tenant, and centralised analytics for network administrators — without running separate infrastructure per clinic.
We implemented this with a schema-per-tenant database strategy, tenant resolution at the edge before any application logic runs, and a role and permission model that scopes every data access to the authenticated tenant. A network administrator sees aggregated, anonymised analytics across all clinics in their network. A clinic practitioner sees only their clinic's data. A patient sees only their own records.
The Evidence Database
The quality of the protocol recommendations is only as good as the evidence database backing them. We built a curation pipeline that ingests peer-reviewed physiotherapy research, clinical guidelines from relevant bodies, and practitioner-contributed protocol data — normalising it into a structured format the recommendation engine can query efficiently.
Each protocol in the database carries provenance metadata: the source, the evidence grade, the patient population it applies to, and any contraindications. The recommendation engine surfaces this metadata alongside each suggestion so practitioners can make informed decisions about which recommendations to follow.
Compliance by Design
Healthcare software cannot treat compliance as a retrofit. Every AI decision in Physiolaxy generates a structured reasoning trace that is stored alongside the clinical record. Every data access event is logged. Patient data is encrypted at rest and in transit. The system is architected to support HIPAA and GDPR requirements, and administrators can export full audit logs for regulatory review at any time.
The clinical AI is explicitly an assistant, not a decision-maker. Every assessment and recommendation is labelled as AI-generated, and the practitioner's override is always available with a note field — creating a clear chain of clinical accountability.
What We Learned
Three lessons that shaped the final architecture:
- Clinicians trust transparency more than accuracy. Early versions prioritised recommendation accuracy at the expense of explainability. Adoption improved significantly when we added visible reasoning to every recommendation — even when the underlying accuracy was unchanged. Clinicians need to understand why, not just what.
- Multi-tenancy is harder than it looks. Schema isolation is straightforward. Ensuring that scheduled jobs, background workers, and cache layers never leak data across tenant boundaries requires discipline at every layer of the stack, not just the database query level.
- The intake UX determines everything. The most sophisticated assessment engine produces poor results if the intake data it receives is low quality. Significant iteration on the intake agent's questioning strategy — how it asks, in what order, with what branching logic — had a greater impact on assessment quality than model improvements.
Frequently Asked Questions
What technology stack is Physiolaxy built on?
The platform is built on a Next.js frontend with a Node.js backend and PostgreSQL for data storage, using a schema-per-tenant multi-tenancy model. The AI assessment and recommendation engines are built on top of Claude's API, with a custom evidence retrieval layer using pgvector for semantic protocol search.
Can Physiolaxy integrate with existing clinic management systems?
Yes. Physiolaxy exposes a REST API for integration with practice management software, EHR systems, and booking platforms. Common integrations include appointment scheduling systems, billing software, and patient communication platforms.
Is Physiolaxy available for other clinical specialities beyond physiotherapy?
The architecture is designed to be domain-adaptable. The assessment engine and multi-tenant infrastructure are generalisable. Contact us if you are interested in applying the platform to a different allied health speciality.
Conclusion
Physiolaxy demonstrates what becomes possible when AI is treated as a first-class architectural component rather than a feature layer. The result is a clinical tool that reduces administrative burden, elevates protocol consistency, and scales from a solo practitioner to a national clinic network — without compromising on the auditability and compliance standards that healthcare demands. This is what we mean by agent-first software at Vantaverse.