Case study · Education
Scaling a learning platform from 4,000 to 90,000 concurrent learners
Exam weeks brought the platform down every term. We rearchitected around the actual traffic shape: quiet for weeks, then a wall.
- Client
- Education technology provider
- Industry
- Education
- Region
- India & Singapore
- Duration
- 8 months
- Team
- 6 engineers
- Delivered
- 2024
The problem
Traffic was extraordinarily spiky — near-idle for weeks, then 20× load during exam periods, and the platform failed every term.
Video delivery was served from the origin, so bandwidth costs scaled linearly with enrolment.
Assessment submissions were lost during peaks, which is the single worst possible failure for an exam platform.
What we did
- 1
Submissions were decoupled behind a durable queue and acknowledged the moment they were persisted, so a slow grader never loses a student’s work.
- 2
Video moved to a CDN with adaptive bitrate, cutting origin bandwidth and improving playback on constrained connections.
- 3
Autoscaling was rebuilt around scheduled pre-warming for known exam windows rather than reactive scaling that arrived too late.
- 4
We added load testing to CI at realistic peak volumes, so capacity regressions are caught before the term starts.
“Three exam cycles without an incident. Before this project that sentence would have been unthinkable.”
Tell us what you need built
Send over the brief, or just a rough description. We read every enquiry ourselves and reply within one working day, usually with questions before a price.