Helpful, not autonomous.
The sector built AI features that act, decide, monitor, and grade. We use AI to help educators author scenarios. That is where we stop.
AI has one job in Meandrix. Authoring.
The one place AI helps.
An educator describes the situation they want to build. They describe the clinical context, the learning objectives, what resources to use, and the kind of decisions they want learners to face. This context prompts the AI to generate an initial scaffold. The educator then edits, refines, validates, and approves the activity before it is deployed.
AI runs once per scenario to generate a draft for educator review. Everything after that initial draft is human work.
Where AI never enters.
After deployment, AI is not in the assessment loop. It does not deliver the activity. It does not mark, score, or see learner work. It does not decide which pathway a learner follows. Those pathways are executed by the LogicPath engine using rules the educator has defined, reviewed, and approved.
The roles of teacher, examiner, and marker stay with the qualified people responsible for them.
Meandrix provides the structure. Clinical and academic judgement remains human.
The scaffold. In plain terms.
What the educator gets back from a single “Build with AI” call. Everything below is a draft for review, not a deployed artefact.
The scenario context
The initial setup the learner sees includes the case, the situation, and any supporting clinical information (vitals, ECG descriptions, handover notes). This includes variant context blocks so the scenario reads coherently regardless of how the learner arrived at the next decision point.
Pathway structure and routing
The pathway movement logic For Encounters and Consults. AI drafts how Q1 options route learners into pathways, and how subsequent decisions move them between pathways (recovery, maintain, deterioration).
Questions, options, and reasoning
Q1 and the per-pathway questions, the response options learners choose between, and the rationale shown for each option after a learner selects it.
Decision feedback and outcomes
The post-pathway evaluation of the learner's decisions, and the final narrative for each pathway.
The line we will not cross. Equally explicit.
What we choose not to do with AI matters as much as what we do.
AI never grades a learner response, determines a score, or decides whether a learner passes or progresses.
AI never chooses which pathway a learner follows. The LogicPath engine executes rules defined by the educator, deterministically.
AI never sees learner submissions, attempts, names, or results.
AI never speaks to a learner. There is no chatbot, no AI tutor, no AI examiner in the assessment.
AI never authors a scenario without educator review. The educator signs off every deployed scenario.
AI never trains on Meandrix data. This is contractually enforced with Anthropic.
Where AI calls go. And what they carry.
When an educator clicks “Build with AI” in Studio, their brief and the structural parameters of the scenario are sent to Anthropic's API. AI returns a complete draft scenario. The educator reviews, edits, and approves before any deployment.
This is a single round trip per scenario. AI is not called again unless the educator chooses to regenerate from scratch.
No learner data is ever included. No data from other scenarios is included. There is no background data sync, no telemetry, no ongoing connection.
AI assist call
What crosses the boundary, and what does not
What we send
- → The educator's brief, including the case context they have written
- → Optional publically available resources they have added (reference text or URLs)
- → Structural parameters including discipline, difficulty, question count, pathway count, options per question
What we do not send
- Learner data such as names, attempts, responses, or anything from sessions
- Other scenarios in the library
- Session records or Sentinel integrity logs
- Anything beyond the single authoring request
Where the line is drawn. And what stays inside it.
Authoring content only
The only data that touches an AI provider is what an educator is actively authoring in Studio. Learner data is never included. No data from other scenarios is included. No background sync, no telemetry.
No training, contractually
Anthropic does not train any of its models on Meandrix data. This is a contractual commitment, not a policy preference.
One provider, named
Anthropic is the only AI provider Meandrix uses. We do not chain through multiple AI providers, route through aggregator APIs, or use models we cannot account for.
Auditable on request
Institutions running an AI governance audit can request a full account of where AI is used, what data flows where, and the contractual posture. Available to institutional administrators.
Want to walk through this with your governance team? Book thirty minutes.
I can walk you through how AI is used in Meandrix, what data flows where, and the contractual posture — in the terms your AI governance framework asks for.
Book the conversation