Design
Safety case lite
Hazard log, mitigations, residual risk, who is accountable.
Healthcare AI Applications
Clinical and operational AI under real constraints: safety, evidence, privacy, and why a demo on MIMIC is not a deployment.
Created by Baljeet Dogra
Course Objectives
- Pick a use case that is operationally real, not a paper recreation.
- Write the evidence bar and the human-in-the-loop before the model.
- Handle identifiers, linkage and logging as first-class design.
- Defend the system to a sceptical clinical safety officer in twenty minutes.
What you will produce
Build
Scoped prototype
One workflow, evals, privacy notes — not a diagnostic oracle.
4 weeks curriculum
Expand a part for the syllabus. Content stays searchable when closed.
01 The setting Week 1
Clinical vs operational. Workflow, liability, why accuracy is not the product.
- Workflows
- Roles
- Liability
- Evidence
02 Data and privacy Week 2
Identifiers, de-identification theatre, audit, retention.
- Identifiers
- Audit
- Retention
- Access
03 Safety Week 3
Hazards, HITL, fail-safe, evals that match harm not leaderboards.
- Hazards
- HITL
- Evals
- Fail-safe
04 Defence Week 4
Prototype plus a twenty-minute safety walkthrough.
- Prototype
- Safety case
- Limits
- Capstone
Who this is for
Engineers in health or life sciences
You are being asked for “an LLM on the notes”. You need a bar.
Informaticians
You know the domain. You need production AI literacy.
Prerequisites
- Python and APIs at course-catalogue level
- Willingness to work with synthetic or public de-identified data only
Not a fit if
- Building a diagnostic product in four weeks
- Anyone expecting real patient data in the lab
Related: AI Security & Privacy · AI Governance & Compliance
Questions
NHS-specific?
UK-aware (clinical safety thinking, IG). The engineering transfers to other health systems.
Will we use real records?
No. Synthetic and public de-identified sets only.
Ready to start?
Four weeks to a scoped healthcare system you can defend. Create an account to enrol.
Enrol now