Agent
Patient Interview
Starts with patient priorities, not clinician assumptions.
CRISP Live
A replayable agent team for clinical reasoning.
CRISP Live shows how I use agents as a reasoning team. Not as one hidden answer, but as a visible process with roles, challenges, artifacts, and review.
The point is simple: if an agent workflow supports healthcare thinking, the reasoning should be inspectable.
- Workshop demo
- Replayable workflow
- Not medical advice
Agent map
challenge
response
artifact
QA
Why it matters
The value is not that agents give an answer. The value is that the process can be inspected.
Healthcare reasoning is full of tradeoffs. A useful agent setup should not hide those tradeoffs behind a polished final paragraph.
CRISP Live makes the reasoning process visible. One agent frames the case. Another checks evidence. Another challenges assumptions. Another handles safety. The final output only becomes useful after the challenge loop has changed it.
That makes it useful for teaching, workshops, and explaining what a practical multi-agent workflow can do.
Agent roles
Each agent has a narrow job.
The system works because the roles are bounded. No single agent is asked to do everything.
Agent
ICF Goals
Turns the patient story into function, activity, participation, and context.
Agent
Clinimetrics
Chooses what to measure, when to reassess, and what threshold matters.
Agent
Evidence Scout
Checks external evidence and makes the implication visible.
Agent
Devil Advocate
Challenges weak assumptions, missing risk, and premature closure.
Agent
Lead Synthesis
Builds one coherent plan after the challenges have been handled.
Agent
QA Audit
Checks safety, evidence traceability, and patient-centered consistency before sign-off.
Replay model
A run is stored as events and artifacts, not hidden state.
SwarmRun
Stores the case run: mode, patient label, created time, and event count.
SwarmEvent
Stores the visible timeline: status, challenge, response, and artifact events.
SwarmReport
Stores the written outputs: reasoning notes, plans, handouts, audits, and reports.
SwarmTrace
Adds structured trace rows for more advanced replay and workflow analysis.
Artifacts
The system produces things a human can review.
That is the part I care about most. A useful agent workflow should leave behind a trace: what was considered, what was challenged, what changed, and what still needs human judgment.
Reasoning map
ICF and HOAC notes
Clinimetrics choices
Safety checks
Challenge memo
Plan and handout
QA audit
Where it fits
CRISP Live is part of my agent workflow layer.
It sits next to Practice Shift and EBPcharlie as another example of turning process into a product surface.
Clinical reasoning workshops
Agent workflow demos
Physiotherapy education
Explaining safety gates
Showing how disagreement changes the final output
Clinical boundary