ZOLLEGE AI

The AI layer under
healthcare education.

A control plane for allied-health training: curriculum generation, a mastery knowledge graph, live classroom devices, and an audited tool gateway — deployed across a campus network that already exists.

275
Active campuses Across 28 states
21
AI agents in production Code-verified, deployed fleet
5
Program lines Dental, medical, and three more

The constraint

Clinical training does not scale by hiring.

Every bottleneck in allied-health education is a supervision bottleneck. One instructor, one bench, one pair of eyes. Software that only stores records does nothing about it — the work is watching, correcting, and deciding what each student needs next.

Perception

Nobody sees the bench

Hands-on technique is assessed from a checklist filled in afterwards, not from what actually happened at the chair.

State

Progress is seat time

Systems record attendance and grades. They do not hold what a given student has actually mastered, or what they should do next.

Reach

Every system is separate

LMS, SIS, externship partners, and compliance records do not talk, so any new capability has to be built once per system.

The platform

One control plane, six surfaces.

Every campus, classroom, and student record runs through the same core. Each surface below is deployed, not a roadmap item.

01

Curriculum pipeline

Three-pass generation from uploaded materials into structured courseware.

02

Knowledge graph

Concepts, prerequisites, and per-student mastery state with gap analysis.

03

MCP gateway

One authenticated endpoint, permission-gated and fully audited.

04

Live classroom devices

Camera fleet streaming to a realtime voice model that sees the bench.

05

Career network

Externship and job matching with scored resumes and mock interviews.

06

Compliance layer

Multi-state accreditation, audits, and document control with AI triage.

Surface 01 · Curriculum pipeline

Protocols in, courseware out.

  • Three-pass generation. Uploaded protocols, checklists, and reference material become modules, assessments, and rubrics.
  • Structured, not prose. Output lands as courseware objects the rest of the platform can reason over, not as generated text.
  • Partner-specific. A health system's own competency definitions drive the pathway, so readiness maps to what they hire on.
  • Canvas-native. Delivered through the LMS already running on every campus.

Surface 02 · Knowledge graph

Mastery, not seat time.

  • Concepts and prerequisites. The subject matter is modelled as a graph, so a gap can be traced to what it depends on.
  • Per-student state. Mastery is tracked against that graph rather than against position in a syllabus.
  • Gap analysis. The next task is chosen from evidence, which is what makes adaptive delivery possible at all.
  • Early intervention. The same state feeds at-risk identification before a student fails, not after.

Surface 04 · Live classroom devices

A model that sees the bench.

  • Camera fleet in the lab. Purpose-built devices stream the working chair, not a lecture podium.
  • Realtime multimodal model. Technique is observed as it happens, with voice guidance back to the student mid-procedure.
  • Supervision that scales. This is the surface that attacks the instructor bottleneck directly rather than working around it.
  • Instructor QA. The same feed supports class quality review and readiness assessment.

Surface 03 · MCP gateway

Build a capability once.

  • One endpoint. Agents reach the LMS, SIS, warehouse, externship partners, and compliance records through a single authenticated gateway.
  • Permission-gated per user. The tool catalogue a caller sees is scoped to their grants, and every call is audited.
  • 41 tools in one session. That is one permission set, not a platform maximum.
  • Why it compounds. Each new agent inherits every integration already built, so the marginal cost of the next one keeps falling.

Surface 05 · Career network

Training is only half of it.

Placement is where an allied-health credential either pays for itself or does not. The career platform is the surface with the most direct line to that outcome.

10,035
Students on the career platform
3,007
Resumes scored by AI
8
Agents in the career stack Of the 21 in production

Distribution

Deployed, not piloted.

This is the part a competitor has to build before the technology matters. The control plane runs across an operating campus network today, which means every surface in this deck has somewhere to be used.

275
Active campuses
28
States
5
Program lines

Campus footprint is concentrated: the four largest metros are Houston, Dallas, Austin, and San Antonio. Density is real, national reach is uneven, and the map on zollege.ai shows it as it is.

Why it compounds

The mechanism, not a projection.

  • Supervision is the binding constraint. Relieving it with perception rather than headcount is what changes the cost curve of a campus.
  • Shared integration surface. Every agent built on the gateway inherits the integrations already there, so capability accrues faster than the work does.
  • Owned distribution. The network is the deployment target and the proving ground, which removes the sales cycle that a vendor would face.
  • The same stack sells outward. What runs the campuses is what a health-system partner buys.

Deliberately stated as mechanism. Unit economics and market sizing are not asserted in this deck — see the note on the next slide.

Basis of figures

Where each number comes from.

FigureSource
275 active campuses
28 states		 HubSpot company records with active_school = True (290), less 15 whose operation_status shows they are not operating. Pulled 28 Aug 2026.
21 AI agents Code-verified agent classes: 13 in zql-ai, 8 in staff-solutions. Excludes regcom triage agents.
41 MCP tools Live tool catalogue for one authenticated session. Permission-gated per user, so not a platform maximum.
10,035 students
3,007 resumes		 staff-solutions row counts: public.students and public.student_resume_ai_reviews.
5 program lines Warehouse school_metrics grouped by program type, 27 Aug 2026.

Not in this deck, and deliberately so: market sizing, unit economics, and student completion or placement rates. The first two are not yet sourced; the third are regulated disclosures for career schools and must be reconciled with accreditor and state filings before external use. See src/data.js in the zollege.ai repository.

What we are building

The infrastructure layer
for clinical training.

Not a courseware vendor and not a point tool. A control plane that sees the bench, holds what each student knows, and reaches every system a campus runs on.

zollege.ai  ·  [email protected]