AI and safety
Helpful intelligence. Human decisions.
QuickMed uses AI as a constrained assistant inside professional workflows — never as an autonomous medical practitioner. Here is exactly where it helps, and exactly where it stops.
Capabilities
Five assistive capabilities, five human checkpoints.
Clinical scribe
Structures consented dictation into symptoms, onset, severity, history, allergies, red flags and a draft summary.
Human control · Doctor reviews, edits and approves.
Diagnosis-aware prompts
Uses the suspected diagnosis to identify missing questions, red flags and possible differentials.
Human control · Doctor confirms the final diagnosis and care plan.
Prescription support
Presents medicine options, dose ranges, contraindications, interactions, allergy alerts and availability signals.
Human control · Doctor selects, changes, rejects or adds items and signs.
Medication plan
Turns the signed prescription and counseling rules into dose timing, food precautions and missed-dose guidance.
Human control · Doctor and pharmacist rules override AI.
Record intelligence
Summarizes longitudinal history, flags possible duplication and prepares consented sharing bundles.
Human control · Authorized clinician verifies; patient controls sharing.
Boundaries
Non-negotiable safety rules.
- AI never independently diagnoses, prescribes, substitutes, cancels or changes a dosage.
- Every clinical suggestion is grounded in approved medicine data, a clinical rule, a guideline reference, or clearly labeled uncertainty.
- Clinician edits, overrides and approvals are logged for audit and safety review.
- Patient-facing instructions use plain language and always offer a route to professional help.
- The website chatbot explains QuickMed only — never diagnosis, treatment or personalized medical advice.
- "…cough started about three days ago, worse at night, no fever that I noticed, I get a rash from penicillin, and I finished a course of something last month…"
Illustrative example with fictional data. AI output is a draft until a clinician reviews, edits and signs.
Trust layer
Every claim, expandable.
Open each principle to see how it is implemented rather than asserted.
Access is granted by purpose, record category, recipient, start time and expiry time. Emergency access reveals only critical information and creates a strict audit event.
Every sensitive access, AI generation, clinician approval, prescription change, export and fulfillment event is logged with actor, role, reason and previous/new values.
Data is encrypted in transit and at rest, and locally cached records are protected with secure device storage.
Clinical AI output carries a review state. Until a clinician approves it, it is labeled a draft everywhere it appears.
Consent in practice
Sharing is scoped, timed and revocable.
Build a consent grant
Choose what a recipient may see, why, and for how long.
Resulting access scope
- Recipient
- Parirenyatwa Clinic
- Purpose
- Referral
- Expires
- In 7 days
- Visible
- Allergies, Prescriptions
- Hidden
- Chronic conditions, Consultations, Lab & imaging reports, Adherence logs
Every view of these records is logged with actor, role, action and timestamp. You can revoke this grant at any time.
Architecture
Where the safeguards live.
Safety is not a single feature — it is distributed across the stack that runs the QuickMed mobile application.
Layer 1
Shared Kotlin core
Domain models, validation, appointment states, prescriptions, medication schedules, consent logic, AI contracts, encryption helpers and offline-sync rules.
Layer 2
Android and iOS clients
Mobile experiences for patients, doctors and pharmacists, optimized for mobile-first and low-bandwidth use.
Layer 3
Web portal
Facility administration, provider verification, record retrieval, pharmacy operations, analytics, support and integration management.
Layer 4
Backend and data
Authentication, role-based access, encrypted storage, notifications, audit logs, AI gateway, APIs and structured health-data exchange.
Layer 5
AI gateway
Prompt templates, model routing, grounding sources, constrained output schemas, safety filters, version logging and human-review states.
See the QuickMed journey with your own team.
Clinics, pharmacies, facilities and pilot partners — walk through the connected workflow, ask hard questions about safety and consent, and tell us where it needs to fit your practice.