Scholarium: Faculty-Governed AI Teaching Assistant

Our mission

To make every course teach the way its field agrees it should, under its professor's authority, with evidence. We rebuild courses into measured, faculty-governed active learning, so every student learns with a master tutor, and every professor holds proof of mastery.

Our vision

To be higher education's operating standard for governed AI teaching.

Faculty govern.

The tutor's conduct, scope, and availability sit under settings your faculty control, and approvals, notes, and source changes are recorded to a ledger.

Institutions own.

Your courses, your data, your records, isolated to your institution on every access path and reached at your own address.

Evidence decides.

Design choices trace to published research and pass graded tests before release.

The problems we address

Built for the questions institutions are actually asking

Assessment validity in the AI era

When generic AI completes assignments, grades stop measuring learning. Our tutor coaches rather than completes. It declines graded work. Concept-level mastery gives faculty independent evidence of what each student can do.

AI adoption without governance

Most institutions choose between restricting tools they cannot govern and permitting tools they cannot defend. A governed teaching assistant is an AI the institution can stand behind: faculty-approved sources, policy-set behavior, faculty actions and model calls recorded.

Faculty trust and faculty workload

The professor holds authority over every source and an off switch, and the platform does the assembly. A course is configured through one structured intake, not built on nights and weekends. The professor's teaching layer remains their own intellectual property.

Outcomes and retention

Course-level failure and withdrawal carry real institutional cost. Concept-level measurement shows faculty where a cohort is weak while there is still time to act, before the examination reveals it.

Evidence of learning for accreditation

Accreditors expect documented, systematic assessment of student learning. A governed teaching assistant produces auditable, concept-level learning evidence as a byproduct of normal teaching.

Data governance and procurement

One platform serves every institution, with each institution's records isolated by an institution key enforced on every access path. The institution's data is its own: exported on request, revocable, not used to train AI models, handled under a data protection addendum.

The platform

One platform. Two surfaces. One spine.

An active course for every student, a console for every professor, and a governance spine running under both. The platform assembles each course from the faculty's own materials and builds the teaching assets around it: figures, video lessons, practice items, reviewed by faculty block by block on the console.

SURFACE 01 · STUDENTS

The active course

The course itself, delivered as active learning: tutored lessons, cases, retrieval practice, spaced review, driven by a personal tutor grounded in the course's approved materials. Every course is mapped to its learning outcomes and the concepts beneath them; the tutor works each student along that map, milestone by milestone, and each student sees where they stand and what to work on next. It coaches toward the answer, cites its sources, and declines graded work.

ACTIVE COURSE · MODULE 3 · GENERATED FROM COURSE MATERIALS

Module 3 · Cardiovascular PharmacologyOutcome: dose and monitor safely in heart failure

CONCEPTS HELD · 7 OF 12

Video lesson · 6 minHalf-life and dosing intervals

Figure · First-order eliminationfaculty-approved

Retrieval practice · 8 itemsspaced review scheduled

YOUR TUTOR · WORKING THIS MODULE'S CONCEPTS

Try it on a new case

COACHING FROM THE COURSE'S OWN MATERIAL · SOURCES SHOWN

SURFACE 02 · FACULTY

The faculty console

Where governance is exercised. Professors review each module block by block, set tutor conduct, and hold the off switch, without creating from a blank page. Per-concept standing for the class, a needs-attention list with its plain reason, calibration counts, and activity show where intervention matters most. The platform computes no grade of its own and writes nothing to the gradebook unless the professor turns on one informational column.

CONSOLE · THIS WEEK

Needs attention · 12 studentsstruggling on clearance after repeated checks

Open list

Standing by concept · class of 84

Pharmacokinetics

Weight-based dosing

Mechanisms of action

Calibration · 9 answered sure and wrong on clearance this week

What students asked the tutor aboutclearance, 31 questions this week

Missed together14 students confuse half-life with clearance

THE SPINE · UNDER BOTH SURFACES

The governance spine

Not a place anyone visits: the properties running under everything. A release gate every change passes before it reaches students. Faculty approvals, notes, and source changes recorded to a ledger. Tutor conduct under faculty settings, with an off switch per course. Data isolated to the institution that owns it.

GOVERNANCE · ACTION LOG

assembled practice set · 8 itemssource: approved bank · trigger: study session

extra practice · 3 items generatedlabelled ungraded · not counted toward standing

request to grade written workrefused: outside its authority

STANDING TEST SUITEPassed before any change goes live

OFF SWITCHPer course · on the professor's console

In the ordinary week

What a week looks like

The platform is built to be used in the ordinary week of a course, not around it. Here is what reaches each seat.

For a student

A tutor inside each module, grounded in the material your professor approved. It coaches toward the answer and shows its sources.
When a concept reads as solid, the tutor offers a new case to try it on. That is how it checks the idea transfers, not just that the answer was remembered.
Practice items drawn from the approved bank, matched to the concepts you hold least well.
Flashcards built from approved check items, and a study guide you can export from the module's approved blocks.
Equations and chemical notation rendered properly, so a formula reads the way it does in your textbook.
Select any passage and ask about it. Save what helps to your notebook.
Spaced review, scheduled per concept. If you opt in, one reminder a day, inside the hours you set.
A clear standing: which concepts you hold, which you do not yet, and where there is not yet enough evidence to say.

For a professor

One digest a week: where the class stands by concept, who needs attention and the plain reason, what students asked the tutor about most, and which concepts they miss together.
A console question box. Ask about this course in plain language and get an answer drawn from the class record, nothing else.
Every setting the tutor obeys, on one page: scope, conduct, availability, the transfer check, and the off switch for the course.
Block-by-block review of each module. Your tick is the approval; the module seal is the formal review. Both are recorded.
Usage and cost for the course, so the model bill is visible before it is a surprise.
An office-hours line the tutor quotes when a question is yours to answer.
No grade computed by the platform. One informational gradebook column exists and stays off unless you turn it on.

For every seat

Three concerns. Three answers.

For students

"Will this actually help me learn, or is it another portal?"

A personal tutor inside the course, available whenever you study, grounded in what your professor approved, and honest with you. It shows which concepts you hold and which you don't yet, coaches you through the gaps, schedules review before you forget, and adapts to how you're doing and what you prefer. It does not do your work for you. It works with you until you can.

For faculty

"Do I keep authority, and does this cost me my nights?"

You govern the sources and the tutor's conduct, and hold the off switch, while the platform does the assembly from materials you already have. You see your class concept by concept: which concepts are shaky, who has stalled, who is struggling after repeated checks, before the exam finds out. Once a week the digest also tells you what students asked the tutor about most and which concepts they miss together, so the next lecture can start where the class actually is. Your teaching voice remains your intellectual property, and it travels with you. The platform computes no grade; the one gradebook column it can write is informational and off unless you turn it on.

For institutions

"Can we defend this to accreditors, counsel, and our own faculty?"

An AI the institution can stand behind: faculty-governed by design, release-gated before students, with faculty actions recorded. Concept-level learning evidence accrues as a byproduct of teaching, ready for accreditation review. Where a class is weak surfaces while there is still time to act. And your data is isolated to your institution on every access path: exported on request, revocable, not used to train AI models.

The pedagogy

An evidence-based spine. Your discipline's frameworks. Control at every level.

The spine is engineering, not opinion

Every course runs on Bloom-aligned outcomes, mastery gates, retrieval practice, spaced review, and concept-level remediation: the mechanisms the learning sciences consistently support, built into the platform itself.

Anchored to your field's own frameworks

Everything discipline-specific is the faculty's choice. Each course anchors to the consensus frameworks of its own field, whether clinical judgment models, case-based learning, safety-competency standards, or the equivalents in your discipline, selected and confirmed by the professor, cited to their sources.

Customizable at three levels

The institution sets policy. The faculty member sets the course: sources, conduct, frameworks, off switch. The professor also sets what the tutor does when a concept reads as solid: the transfer check is a switch, on by default, off in one click. The student sets preferences the evidence says are theirs to hold. Every level visible to the one above it, nothing hidden from faculty.

How mastery is measured

Course objectives→ Module & lesson objectives→ Concepts, from your frameworks→ Assessments mapped to concepts→ The mastery profile

Mastery is measured against concepts and frameworks, not pages and chapters. Every objective traces to the one above it, every assessment item maps to a concept, and the profile shows which concepts each student holds and says when there is not yet enough evidence.

Governance & trust

Why institutions choose Scholarium

Inside your LMS.

Scholarium runs as an LTI 1.3 Advantage tool. Students and faculty enter from the course page they already use. Rosters come from the LMS. Grade writeback is off unless the professor turns on the one informational column.

A release gate.

A judged suite of tutor conversations runs before any change reaches a student. Live courses are re-tested nightly.

Safety content stays word for word.

In courses where a wrong number is a clinical error, the professor's safety content is served verbatim, and a dose the material does not contain is refused rather than computed.

A ledger.

Faculty approvals, notes, source changes and tutor settings are recorded with who and when. Course content is versioned. A module serves live after a sealed walk or a recorded override, and not before.

Your data, in your institution.

Records are isolated to your institution on each access path and reached at your own address. Not used to train AI models.

A written retention schedule.

Each class of record has a stated retention period. Export on request. A student can see their own data page. Records are deleted on the institution's instruction.

Reporting floors.

Inside the course, faculty see their class at any size. Outside it, nothing is reported below a group of ten.

Accessibility.

Built and tested against WCAG 2.1 AA in the deploy gate. Videos carry captions and a transcript. Figures carry written descriptions. What is tested, and what is not yet done, is on the accessibility statement.

Model calls metered.

Each model call is counted per course and shown to the professor.

Evidence

The numbers we build from

2^σ

The benchmark · one-on-one tutoring

Students tutored one-on-one under mastery learning outperform conventional classrooms by about two standard deviations.

BLOOM · EDUCATIONAL RESEARCHER · 1984

0.73–1.3^SD

The trial · a purpose-built AI tutor

In a randomized controlled trial at Harvard (N = 194), students learning with a purpose-built AI tutor outperformed an expertly run active-learning class by 0.73–1.3 standard deviations, with higher engagement, in less time.

KESTIN ET AL. · SCIENTIFIC REPORTS · 2025

How we measure

Outcomes on this platform are measured, not claimed. Each pilot runs a baseline, a midline and an endline survey that the professor approves before it goes to students. Concept-level standing accrues through the term. Exam results can be imported per student on the institution's terms and read against that standing. Nothing is reported outside the course below a group of ten. Results are written up with the faculty who ran the course, and the protocol is recorded in a pre-registration ledger.

About Scholarium

Scholarium builds the faculty-governed AI teaching assistant for university courses. We build, test, and operate it with the faculty who own the courses, and every release passes a judged test suite before students see it. Our first courses are built with university faculty in nursing and medicine, at an institution accredited by a U.S. regional accreditor, disciplines in which the cost of a content error is clinical and the standard for source fidelity is strict. The same architecture serves any discipline's frameworks: business, engineering, the sciences, the humanities.

Scholarium was founded by Faisal Darwiche, NP, Founder & CEO, who leads the company's pedagogy and platform. MSN, AANP board-certified, 27 years in clinical practice, founder of three medical practices.

See the platform working, not described.

Explore it yourself, on your own time: work a module as a student, run the console as the professor. No call scheduled, no one following up. When you want it on your campus, the form below reaches the founding team directly.

Explore the platform

Replies come from our founding team.