An EdTech startup wanted to compete with the unicorns on tier-2 pricing — without sacrificing class quality. We delivered an LMS with WebRTC live classes, HLS-streamed recordings, doubt-clearing, mock tests, and parent visibility — scaling to 60,000 students at a fraction of the unicorn-cost-per-class.
The client had a strong content team and 18 months of recorded content trapped on Google Drive. They were running live classes on free Zoom, payments via Razorpay links pasted into WhatsApp, and 'doubt clearing' via voice notes. The system fell over at 4,000 students. They needed a real platform — fast — to capture the next admission cycle.
Four apps + a video-streaming spine. The non-obvious choice was building our own WebRTC-based live class engine instead of wrapping Zoom or 100ms — saved Rs 18-22 lakhs/month at scale.
Class schedule, live class join, recorded content with bookmarks + speed control, doubt-ask (text + photo + voice), mock tests with instant scoring, leaderboard, mentor assignment, parent-link, peer chat (moderated).
Class roster, take live class (with whiteboard + screen-share + breakout), record session, upload supplementary content, grade mocks, answer doubts, MIS dashboard, payout view.
Child's attendance, performance, mock test scores, fees, teacher messages. This single feature cut cancellations 38%.
Course builder (modules, sub-modules, lessons), schedule manager, teacher allocation, fee + scholarship engine, curriculum dashboard, retention cohorts, support desk.
Live class via WebRTC SFU (mediasoup) for <200ms latency in classes <200 students. Auto-record to S3, transcode to HLS for VOD. Adaptive bitrate for tier-3 mobile networks. CDN-edge cached.
We picked the stack for fit, performance under realistic load, and operational simplicity. Here is the breakdown:
mediasoup over 100ms / Agora for live classes: we self-hosted an SFU. Cost-per-minute is 1/8th of managed competitors. The trade: we own the on-call; the team learnt mediasoup quickly enough that it was worth it. At 500,000 minutes/month you cannot afford managed.
HLS via MediaConvert + CloudFront for VOD: recordings are transcoded to multi-bitrate HLS, served via CloudFront with regional caches. Tier-3 students on 3G can stream at 240p without buffering; metro students get 720p.
Mocks engine in Postgres: a mock test is a JSON template; student responses are JSONB; scoring is a SQL function. Adding a new exam pattern is a JSON definition, not code.
Live class quality on tier-3 networks.
Indian education's biggest market is in tier-2 and tier-3 cities — exactly where networks are worst. A live class that buffers loses students permanently; one bad week of classes and they unsubscribe.
We attacked this in three ways: (a) adaptive bitrate with simulcast — every student gets the bitrate their network can handle; (b) intelligent reconnect — if a student drops, they rejoin within 4 seconds without losing context; (c) auto-record + HLS fallback — if live fails, the student gets the recording within 90 seconds.
Result: 92% live uptime measured at the student edge (not the server edge — that is the metric that matters). 8% who hit issues land on a 90-second-delayed HLS fallback.
We launched on a single batch — 800 hand-recruited Class-12 NEET students — for one term. Three months of intense feedback loop. The app changed shape twice in that period; every change went into v1 release.
Scale-up was admission-cycle-driven. April–June we ramped from 800 to 14,000. Sept–Nov we hit 60,000. Infra scaled fine because we'd loadtested at 100,000.
“Parent visibility was the unlock we did not see coming. Our content was always good; the platform just made the parent feel they were getting their money's worth — and renewals followed.”
Co-Founder
EdTech Platform · NEET / JEE Prep
Build the parent app in v1, not v2. We launched without it; cancellations were our biggest leak for the first term. Once we shipped it, the rate dropped 38%. Parents are the buyer; treating them as a non-user was the strategic miss.
Live classes, recorded VOD, assessments, parent visibility — these are not features, they are infrastructure. We've shipped EdTech platforms scaling to six-figure user bases on Indian-network realities. Get a costed roadmap.
Get a Free Consultation
