PsychApp v0.2: clinician-supervised by design
PsychApp v0.2 reframes the project as a preclinical research proof of concept. It separates population modelling, a participant-level digital twin and longitudinal state updates while keeping diagnosis, treatment decisions and emergency use outside the product boundary.
Architecture assembled in the repository
- PWA: participant and professional views built with React and Vite.
- Gateway contracts: a serverless reference layer for authentication and consent boundaries.
- Model service: a Python POC with coordinated agents, ephemeral storage and mock adapters.
- Data layer: a versioned Supabase schema with RLS, consent, traceability and human-review concepts.
- Research assets: reproducible population-baseline and QLoRA configuration material.
Safety is a separate execution path
Safety analysis runs on the original Spanish text. Translation and voice are optional helpers and cannot replace the source or determine risk. Predictions remain probabilistic support for review, not orders.
The public build uses synthetic content. It is not a medical device, diagnostic service, treatment system or crisis tool, and it does not establish clinical efficacy.
Repository capability is not deployment proof
The Render host can serve the PWA and blog, but the gateway, Python model service and Supabase schema are separate modules. Their presence in Git does not mean they are connected to the public host or authorized for real patient data.
Deployment acceptance requires the public health response to identify release and commit, plus verification that the custom domain serves that exact build. Until that check is complete, repository state and live state must be described separately.
Employer-facing evidence
The value of this version is the engineering reasoning it exposes: explicit data boundaries, replaceable adapters, review queues, versioned models, reproducible research assets and a deployment contract that can be inspected instead of assumed.