AI Assignments That Meet Students Inside Your LMS
Role-play simulations, staged scenarios, oral exams, graded write-ups, and prompt-engineering labs — with evidence-cited AI grading your instructors review before it ever reaches the gradebook. Delivered through LTI 1.3 in any compliant LMS.
Students can speak or type · Ungraded practice mode · Instructor review on every grade
Try LTISim in Action
This is a live role-play simulation — the same kind students run inside their LMS. Step into the scenario below and talk to the AI character yourself.
Frustrated Patient Encounter
You are the healthcare professional. Margaret has been waiting two weeks for biopsy results and is anxious and upset. Acknowledge her concerns, show empathy, and deliver her results clearly.
In the product, replies stream in live, students can speak their messages, and every attempt can start as ungraded practice.
One platform. Five ways to learn with AI.
From role-play to oral exams — every assignment type runs through the same rubric grading, evidence-cited feedback, and instructor review pipeline.
Simulations
Students hold a multi-turn conversation with an instructor-defined AI character. The full transcript is captured and graded against your rubric — so assessment reflects how students actually handled the interaction, not just whether they reached a conclusion.
- Instructor-authored persona, scenario, and learning objectives
- Students can speak their responses or type — spoken input is transcribed and editable before sending
- Grounded in your course materials and guardrails
- Full transcript graded against a rubric you control
A nursing student conducts an intake interview with a patient persona — graded on active listening, clinical reasoning, and bedside manner.
Every type, same pipeline
Built for the instructor with 90 other things to do.
Author an assignment in minutes, rehearse the grading before students see it, and read the whole class at a glance.
Author in minutes, not evenings
Describe the assignment in plain language and get a complete draft — persona, scenario, welcome message, and a full rubric — in about a minute. Then shape it: refine any field with AI help, or start from a curated template instead.
- Two-sentence description → complete draft assignment, ready to edit
- Per-field AI refine and an AI rubric review (vague levels, overlap, coverage gaps)
- Test-grade a pasted transcript before publishing — see exactly how the rubric behaves first
- Template gallery, one-click duplicate, copy-to-course, import/export
AI: They said it's a blood thinner? I'm worried about…
Read the class at a glance
Scores tell you who struggled. Criterion-level analytics tell you what to teach next: every assignment aggregates by rubric criterion, so patterns across the class are visible — not buried in individual transcripts.
- Per-criterion class view — see which skills the class is strong or weak on
- Score distributions and a per-student completion roster
- A cross-assignment review queue with keyboard navigation
- CSV export, and per-criterion improvement tracking across student attempts
We measure the
grader.
An AI grade is only useful if it's consistent and explainable. We treat our evaluator as something to be tested, not trusted on faith — every change runs against a versioned validation suite, and we write down what it shows, limits included.
What we've measured so far
Reproducible scores
In our latest validation run, re-grading each test transcript produced identical scores every time — no run-to-run drift at the low-temperature setting we run in production.
Correct task attribution
On a controlled test set, the grader correctly identified which prompt each task was attempted in — every pair — including out-of-order attempts, clarifying questions, and tasks the student never attempted.
Explained, not asserted
Each criterion score comes with the grader's written, rubric-level reasoning — so an instructor can see why a grade landed where it did, and change it before it reaches the student.
Figures from our internal validation suite — reproducible scripts run against AWS Bedrock, documented per run. Measured on controlled test transcripts, not yet on graded student work at scale. Methodology available on request.
And it's built into the product
Evidence-cited feedback
The grader cites the exact messages behind each criterion score — so "why" is never a mystery, for the instructor or the student.
Confidence-gated release
The grader reports its own confidence per criterion. Scores can release immediately when it’s sure — and hold for instructor review when it isn’t.
Blind-grading agreement
Instructors can grade a session without seeing the AI draft; the platform records both and computes human–AI agreement on real coursework.
Fail-safe grading
If the evaluator fails, it says so and retries — it never fabricates a score. Grading also runs on a separate, stronger model than the conversation.
The next question needs your classroom.
Consistency and attribution we can measure ourselves. Agreement with expert human graders — the metric that matters most for high-stakes assessment — can only be measured on real coursework graded by real instructors. That study is what we're building, and why we're recruiting pilot institutions.
Blind-grading agreement measurement isn't a promise — it's shipped. Pilot partners run it on their own rubrics from day one.
Partner on the researchWhat a research partnership looks like
- →Run assignments in your own LMS, on your own rubrics
- →Compare AI scores against your graders' — agreement, reliability, and where they diverge
- →Co-author what we find, including where the AI falls short
The Platform
Purpose-built for higher education — every feature designed around instructor needs and institutional requirements.
LTI 1.3 Native Integration
Launches directly from any LTI 1.3 compliant LMS. Students stay in their course. No separate platform to manage.
Institutional Controls
Every institution runs in its own isolated tenant, with an audit log of consequential actions, AI usage and cost tracking with optional monthly limits, and a grading-agreement dashboard for measuring the AI against your own graders.
Speak or Type
Students can speak their side of the conversation — transcribed in the browser and editable before sending. Typing is always available; voice is never required.
Feels Live
Replies stream in as the AI responds — the persona answers like a conversation partner, not a loading spinner.
Practice Mode
Instructors can allow ungraded rehearsal runs with real AI feedback. Practice never touches the gradebook and never consumes a graded attempt.
Personas That Stay in Character
Students interact with AI that stays in character, references course materials, and responds dynamically to student decisions within pedagogically structured guardrails.
Built for Experiential Learning Across Disciplines
AI-powered simulations that meet learners where the stakes are highest.
AI Literacy & Prompt Engineering
Students across disciplines learn to use LLMs responsibly through graded prompting tasks — every prompt, iteration, and final artifact captured for review.
Healthcare Education
Nursing and medical students work through staged encounters — intake, examination, care plan — then write the SOAP note, graded together with the conversation it documents.
Business & Management
Students navigate negotiations, stakeholder meetings, and leadership challenges with AI personas — then defend their decisions in an AI-led debrief after the case.
Social Work & Counseling
Practice intake interviews, crisis intervention, and client communication in safe, repeatable simulation environments.
Teacher Preparation
Pre-service teachers practice parent-teacher conferences, classroom management conversations, and IEP meetings — with oral-exam-style defenses of their instructional choices.
Compliance & Professional Training
Workforce learners practice regulatory scenarios, customer interactions, and safety protocols with immediate AI-driven feedback.
Straight answers.
We're in pilot-partnership mode rather than self-serve signup. Each pilot is scoped together with the institution — typically starting with one course and one instructor — and pricing is being defined with early partners rather than published as a price list. If that sounds workable, reach out below.
Something else? Ask us directly — or read the Privacy and Security pages.
Partner With Us
We're seeking institutional partners for research collaborations. If you're interested in bringing AI-powered simulations to your learners, we'd love to hear from you.