Education Software Development

The edtech graveyard is full of beautifully built platforms with excellent content and terrible completion rates. The engineering was never the problem. The problem is that a learner who stops opening the app in week three is indistinguishable, to most software, from one who is doing fine — right up until the refund request arrives.

Engagement is an operational system, not a feature

Self-paced online courses complete at rates in the single digits. Cohort-based programmes with a live component and real accountability complete far better, and the reason is not the video quality. It is that somebody notices when you disappear.

So we build the noticing. A per-learner risk score that combines login recency, session depth, assignment submission and quiz performance — not a vanity dashboard, but a queue that tells a counsellor which eleven learners to call today, ranked, with the reason attached. A nudge ladder that escalates from an in-app prompt to WhatsApp to a human. Cohort-level alerts when a specific module causes a drop-off spike, 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 make it visible early enough to act on.

Proctoring: the honest version

Nobody has solved remote proctoring, and any vendor claiming otherwise is selling you confidence rather than integrity. What is achievable, and what we build, is raising the cost of cheating and producing evidence a human can adjudicate.

That means browser lockdown and focus-loss detection, randomised question banks so no two candidates see the same paper, per-question timing that flags a candidate answering a hard question implausibly fast, webcam snapshots and audio anomaly detection where the stakes justify it, and — the part that actually matters — a review queue where a human makes the final call, always. We do not auto-fail on an algorithm's say-so. For genuinely high-stakes certification, we will tell you that a physical test centre is the only real answer, and 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, multiple-entry and multiple-exit programmes, with the Academic Bank of Credits holding credits that are portable between institutions.

The engineering consequence is specific 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 an accumulating credit balance against a requirement set, where a learner may exit with a certificate after one year, return three years later, and have their prior credits still count. If your data model assumes a student belongs to a batch that progresses through semesters together, you will be rewriting it. We model credits as first-class transferable entities from the outset, because retrofitting that is a rewrite of the academic core.

Video, which is where the money goes

Bandwidth is the largest running cost in most edtech products, and the naive implementation is brutally expensive. Adaptive bitrate streaming so a learner on a 3G connection in a small town gets a watchable stream rather than a buffering wheel. Signed, expiring URLs so your content is not re-uploaded to Telegram within a week of launch. Aggressive CDN caching. Download-for-offline on mobile, because that is how a great deal of India actually consumes learning content, on a phone, on a commute, with no signal.