MD-assistant
A Next.js + Firebase dashboard where medical staff transcribe notes, check spelling, query patient records, and draft clinical documents entirely in natural language. No forms, no navigation — just conversation.
A conversational layer over clinical work. Staff say what they need — transcribe a note, check a spelling, pull a patient record, draft a letter — and the AI reads the intent and the clinical terms, checks them against the Firestore schema, and acts. Patient data stays protected, and the fatigue of endless forms disappears.
What it solves
- Entering medical notes is slow and scattered across screens, and manual spelling and formatting invite errors.
- Pulling a patient record means digging through deep hierarchies; non-technical staff struggle with database search.
- Drafting referral letters and treatment summaries means repetitive typing and manual compliance checks.
Impact
Speak your intent, not forms
Intent → instant Firestore check
Clinical term correction built in
Architecture
Data flow
- Doctor/staff voice or text input
- Vercel AI SDK parses intent + corrects clinical spelling
- Zod schema validation
- Firebase role + permission check
- Firestore transaction (create/update/read)
Patient data immutable, audit-logged
- Real-time UI sync + toast confirmation
Engineering decisions
v0.app-driven UI + GitHub source-of-truth for logic
Visual and component changes live in v0.app and auto-sync here; business logic, validation, and auth stay in this repo, ensuring UX agility without losing code control.
AI-first input pipeline with clinical safety checks
Staff input → LLM parses intent and corrects clinical spelling → schema validation against Firestore structure → role-based access check → atomic transaction to patient record. No client-side bypass.
Radix UI + Tailwind for clinical-grade accessible UI
All inputs, dialogs, and navigation are WCAG-compliant and keyboard-navigable; medical staff may need one-handed or eyes-free operation during rounds.
Gallery
