Education
Education software has a load profile unlike almost anything else: near-idle for weeks, then every student in the institution submitting within the same twenty minutes because the deadline is the same for all of them.
The load is synchronised, and that changes everything
Commercial traffic arrives spread across a day. Academic traffic is deliberately synchronised — enrolment opens, a deadline closes, results publish. Capacity planned against averages will be wrong by more than an order of magnitude at exactly the moment failure is most visible and least forgivable.
The practical consequences are unglamorous: submission has to be durable and idempotent so a student retrying on a failing connection does not lose work or duplicate it, and the queue behind it must accept the spike even when processing lags. A submission lost at a deadline is an academic dispute, not a support ticket.
Where we work
Learning platforms
Course delivery, assessment and progress tracking, including the offline and low-bandwidth cases.
Student information systems
Enrolment, records and timetabling, integrated with the finance and identity systems already in place.
Assessment and integrity
Submission, grading workflows and plagiarism handling, with an appeals trail that holds up.
Institutional integration
LTI, SIS and identity federation, so a new tool does not mean another set of credentials.
Learning analytics
Progress and at-risk indicators built with explicit care about what a prediction is allowed to trigger.
Accessibility remediation
Bringing existing platforms up to standard, prioritised by which students are currently excluded.
Student data deserves more caution than it usually gets
Education data is disproportionately about minors, and the sector has a pattern of collecting engagement telemetry far beyond what any pedagogical purpose requires. The safest position is to collect less, and to be able to explain the purpose of every field you do collect.
Where analytics identify students as at risk, the design question is not only accuracy but consequence. A model that quietly influences how a student is treated, without a human decision and a route to challenge it, is a governance problem regardless of how good the model is.
Data minimisation as a design rule — a purpose for every field collected
WCAG 2.2 AA verified with assistive technology; content authoring included
Durable, idempotent submission so a deadline retry cannot lose work
Predictions surfaced to humans with reasons, never acting autonomously
Retention and deletion aligned to institutional policy and law
Common questions
Yes — LTI for tool integration, plus whatever the platform exposes for rosters and grades. The friction is usually in grade passback and identity mapping rather than in launching the tool, so we would test those two paths early rather than at the end.
Keep exploring.
Ready to start your project?
Get a free technical discovery call. We'll map the right team, stack, and timeline to match your goals — no obligation.