Built against the standards institutions are audited on
The problem
Medical training is one continuum. The software never is.
A student learns on one bank, a resident on another, a physician chases CME on a third. Nothing carries forward. The program director rebuilds milestone evidence by hand every cycle, and the registrar reconciles rosters by CSV. The learning is continuous; only the record is broken.
The record dies at graduation
Four years of calibrated performance data becomes a transcript line. The residency program starts from zero, and so does the learner.
AI is bolted on, not built in
A chat box beside a static bank is not adaptivity. If the model cannot see the blueprint, the psychometrics, and the schedule, it cannot plan a day of study.
Procurement stalls on data
FERPA, PHI in case content, and where the model sends prompts. Products that cannot answer these lose the deal regardless of the feature list.
The platform
Six systems that share one record
Each of these exists elsewhere as a separate purchase. Here they read and write the same learner record, which is what makes the daily loop coherent.
Adaptive QBank
A daily queue built from your blueprint, not a generic bank. Item selection is calibrated against real psychometrics and explains itself — learners see why each question was chosen.
Spaced repetition
Every miss becomes a card automatically. Reviews are scheduled around duty hours and rotation load, so the queue respects clinical time instead of ignoring it.
Clinical case simulation
Branching cases with an AI tutor that probes reasoning, then scores against a deterministic rubric before any model judgement is applied.
Authoring & psychometrics
Faculty draft with an AI assistant, a deterministic flaw linter catches item-writing violations, and published items carry live difficulty and discrimination statistics.
Cohorts & curriculum
Blueprint-true heatmaps show where a cohort is thin before the exam does. Curriculum mapping ties every objective to the evidence that it was actually taught.
Milestones & CME
ACGME sub-competency evidence accumulates from real learner work, then exports as a CCC review packet. The CME data model is built in from day one.
Who it serves
Three audiences, one system of record
The institution is the customer, but the product has to earn its place with the learner every single day. Both things are true, and the design follows from holding them together.
Students, residents, physicians
- One record that survives graduation and program transfer
- A 25-minute post-call session that is actually worth 25 minutes
- Board-readiness forecast with a confidence interval, not a vibe
- Full offline practice with conflict-free sync
Faculty and program directors
- Objective to published item in under fifteen minutes
- Cohort heatmaps that point at the remediation, not just the gap
- Milestone evidence assembled continuously, not the week before CCC
- Every AI draft badged and human-signed before it counts
Registrars, IT, CME offices
- Nightly SIS roster sync and SSO your IT team already runs
- Tenant isolation enforced twice — middleware and row-level security
- Deploy in our cloud or entirely inside your own Azure tenant
- Every action in an immutable audit log, exportable as a binder
Trust
Data sovereignty is the feature
Every claim below is enforced by a technical control, not a policy document. That distinction is the entire reason procurement moves.
Your data never trains a model
Zero-data-retention terms with our model provider, contractually. Learner PII never reaches a model, a log, or an index — pseudonymisation happens before the gateway, not after.
Isolation you can prove
A failed cross-tenant isolation test blocks the deploy. Not a warning, not a ticket — the pipeline stops. Two seeded organisations, zero permitted reads across the boundary.
AI drafts, humans sign
Models draft items, tag content, and explain misses. No model publishes an item, rates a milestone, or issues a credit. Those require a human identity, and the audit log records it.
The shift
The way it’s done. The way it should be.
AI drafts. Statistics flag. Only a named human publishes, rates and certifies. That line is load-bearing for FERPA, ACGME and ACCME — so it is enforced in the schema, not in a policy document.
Build principle 3 — from the VeloLibrary specification