Discovery session with a higher-education programme lead.
Grading assistance pass awaiting educator sign-off.
A Canadian university — first-year calculus misconception pilot
The problem: Four hundred first-years arrived each September with the same chain-rule misunderstanding. Faculty spent the first three weeks re-teaching what should have been caught earlier. A previous vendor chatbot had answered confidently — and incorrectly — when students asked about implicit differentiation.
What we built: An intelligent tutoring flow with retrieval-augmented tutoring tied to the department's approved problem bank. Misconception detection flagged recurring error patterns; hint ladders escalated only after a student attempted a structured step. Every template passed human-in-the-loop review before release. LMS integration layer connected to the existing Canvas instance.
Educator role: Three senior instructors reviewed all hint text, overrode model suggestions weekly, and signed off on each formative assessment module. Model QA and guardrails blocked off-syllabus answers.
Illustrative outcome: Repeat chain-rule errors on the next quiz dropped by roughly 23% compared to the prior cohort — past performance under specific teaching conditions, not a promise for your campus.
A national training provider — accessibility and inclusive learning
The problem: Screen-reader users could not complete a legacy quiz module. Adjustable pacing was absent. An accessibility audit surfaced gaps only weeks before a provincial compliance review.
What we built: An accessibility feature set with caption pipelines, keyboard navigation and screen-reader-tested item rendering. Personalised learning pacing options respected learner profiles without hiding progress from instructors. Full LMS integration documentation covered student-data privacy under PIPEDA.
Educator role: Instructional designers validated every inclusive-learning feature against real learner scenarios. Training providers retained editorial control over all content generation outputs.
Illustrative outcome: Compliance review passed; learner completion rates for assistive-technology users improved in the following term. Your context will differ.
An Ontario school district — K-12 formative assessment support
The problem: Elementary teachers spent Sunday evenings on rubric comments that said the same thing forty different ways. Automated feedback tools from a generic vendor ignored the district curriculum and occasionally suggested answers outright.
What we built: Formative assessment support with rubric-aware comment drafts, question generation for practice sets and academic integrity safeguards that refused to complete assignments. Grading assistance passes queued for teacher approval — nothing reached students without educator sign-off.
Educator role: A pilot group of twelve teachers edited every comment template. District learning design leads owned the knowledge tracing configuration.
Illustrative outcome: Self-reported marking time fell by an average of 4.2 hours per week among pilot teachers. District-wide rollout is ongoing; figures reflect the pilot only.
A college continuing-education unit — content and curriculum generation
The problem: Micro-credential modules needed refreshing twice a year. Subject-matter experts were stretched thin; first drafts took months.
What we built: A content generation pipeline from first draft to reviewed lesson — curriculum-aligned readings, worked examples and formative assessment items. Version control and audit trails satisfied the college's data governance committee. Learner modelling hooks logged which items needed revision after pilot delivery.
Educator role: Programme leads approved every module before enrolment opened. Model QA red-teamed outputs for factual drift and bias.
Illustrative outcome: Draft-to-approved cycle shortened from eleven weeks to five for two pilot programmes. Scalability depends on subject complexity and reviewer availability.