Education Software Development

Many edtech platforms are beautifully built and have excellent content, but very few learners finish. The engineering was never the problem. The problem is that a learner who stops opening the app in week three looks, to most software, the same as one who is doing fine. This lasts until the refund request arrives.

Engagement is an operational system, not a feature

Self-paced online courses are finished by only a few learners. Cohort-based programmes with a live part and real accountability do far better, and the reason is not the video quality. It is that somebody notices when you disappear.

So we build the noticing. Each learner gets a risk score that combines how recently they logged in, how long they stay, assignments handed in and quiz results. It is not a show-off dashboard. It is a list that tells a counsellor which eleven learners to call today, ranked, with the reason attached. Reminders step up from an in-app prompt to WhatsApp to a human. Group alerts warn when a specific module makes many learners quit, because that is a content problem and it will keep happening to every future cohort until somebody fixes it. Completion is an operations problem, and software's job is to show it early enough to act on.

Proctoring: the honest version

Nobody has solved remote proctoring, and any vendor who says otherwise is selling you confidence and not honesty. What can be done, and what we build, is making cheating harder and giving a human evidence to judge.

That means browser lockdown and detecting when the candidate leaves the test, random question banks so no two candidates see the same paper, per-question timing that flags a candidate who answers a hard question too fast, webcam snapshots and detection of odd audio where the stakes are high, and, most important, a review list where a human makes the final call, always. We do not auto-fail on an algorithm's say-so. For truly high-stakes certification, we will tell you that a physical test centre is the only real answer. We would rather say that than sell you a proctoring module that will not survive its first legal challenge.

NEP, and what it means for your data model

The National Education Policy pushes towards credit-based, multidisciplinary programmes where students can join and leave at several points. The Academic Bank of Credits holds credits that can move between institutions.

This has a clear effect on the build, and it is easy to get wrong: a student's programme is no longer a fixed path with a start and an end. It is a growing credit balance against a list of requirements. A learner may leave with a certificate after one year, return three years later, and have their earlier credits still count. If your data model assumes a student belongs to a batch that moves through semesters together, you will be rewriting it. We build credits as core items that can be transferred from the start, because adding that later means a rewrite of the academic core.

Video, which is where the money goes

Bandwidth is the biggest running cost in most edtech products, and the simple way of doing it is very expensive. Adaptive bitrate streaming gives a learner on a 3G connection in a small town a watchable stream instead of a buffering wheel. Signed, expiring URLs mean your content is not re-uploaded to Telegram within a week of launch. Strong CDN caching. Download-for-offline on mobile, because that is how a great deal of India really learns, on a phone, on a commute, with no signal.