Our virtual classroom software development team builds LTI-native classrooms for your LMS, with live captions built in. Attendance and engagement data land in your gradebook from the first class. It's the stack we built for BrainCert (the platform has since delivered 500M+ real-time classroom minutes) and Scholarly (up to 2,000 students in one live class). A virtual classroom in production starts from $58,000.
We built the WebRTC + HTML5 LMS: 10 datacenters, 99.995% uptime.
Live test-prep classes in Sydney; AWS's most innovative EdTech in APAC.

Cloud-based virtual classroom with collaborative whiteboard, HD video, social integration, and white-label API. Perfect for online training and webinars.
Read the case study →.avif)
Virtual classroom that defeats firewalls with WebRTC tutoring.
Read the case study →Most virtual classroom “solutions” are a generic LMS (Moodle, Canvas, Brightspace, Blackboard) with a Zoom or Teams session embedded as an iframe. It works for hour one. It breaks the moment you need attendance tracking that actually posts to the gradebook, AI captions students can search, in-context replay, or a proctored exam. Below: where a custom EdTech video platform earns the rebuild.
Right when: your classroom should be part of your product, not someone else's iframe. From day one. At any size.
Reasonable for: single-class pilot, validating product-market fit, no compliance / gradebook / AI requirements yet.
A virtual classroom isn't a video call — it's a learning event with attendance, engagement, AI assistance, and a recording that has to land in the right student's history. Skip any of those and the LMS goes blind, instructors lose minutes to admin, and the replay nobody can find is the replay that doesn't help anyone.
Scheduling stays in your LMS (Moodle, Canvas, Brightspace, custom). An LTI 1.3 launch hands authenticated students and roster context to the classroom runtime; calendars sync via CalDAV + iCal.
A LiveKit 1.x or mediasoup 3.16 SFU carries host and student WebRTC streams at sub-200ms latency. Breakout rooms, raise-hand, whiteboard, polls, and annotated screen-share run in your branded UI.
Whisper Large-v3 or Deepgram Nova-3 for live captions; SeamlessM4T or DeepL Voice for translation tracks; custom engagement models alert the instructor. AI runs in parallel, so the stream is never blocked.
CMAF / LL-HLS packaging produces replay with chapter markers. The transcript is indexed in pgvector or Elasticsearch, so students search a lecture by phrase and jump to the timestamp. Recordings are encrypted at rest.
xAPI / cmi5 statements post every learning event to the LMS: join and leave times, raise-hands, polls, quiz scores, engaged minutes. An auto-generated summary goes to instructor and students.
Every layer is a deliberate choice for a learning workload, not a generic video call workload. The list below is what we deploy in production today — across higher ed, K-12, corporate training, and professional certification platforms.

Tell us what you want to launch and who it's for. You get back the scope split into what ships first and what waits, the stack we'd pick with the reasoning, and the questions that decide the budget.
Get it free →
Inherited a codebase nobody can explain? We read it and tell you three things: what's worth keeping, what fails under real load, and what a security or compliance review would catch.
Get it free →
Send us how your product works today: what carries the media, where the data lands, and what you've already built. You get back the architecture we'd build instead, what it costs to run at your volume, and which parts you shouldn't build at all.
Get it free →
It works on your machine and falls apart for real users. We put your product through the conditions that break it — weak networks, more people than the demo had, the session that has to be recorded — and hand you each failure with the place in the stack that caused it.
Get it free →Via LTI 1.3 + LTI Advantage. Students launch the classroom from inside the LMS with their authenticated session, the roster + role mapping happens automatically, and every learning event (join, leave, raise-hand, poll answer, quiz score, engaged minutes) posts back to the LMS via xAPI / cmi5 / SCORM 2004. Custom plugins also available for Canvas, Moodle, Brightspace, Blackboard, and BrainCert when LTI isn't enough. If the LMS itself is what needs building rather than integrating with, that is custom LMS development.
Zoom and Teams are general conferencing tools wrapped in an iframe — they don't know they're inside an LMS. Attendance doesn't post back, the recording lives in Zoom's cloud (not your archive), AI captions are vendor-locked, breakout-room state never reaches the gradebook. A custom EdTech video platform treats the LMS as the source of truth and the classroom as a learning event, not a meeting.
Yes. Live transcripts (Whisper Large-v3 or Deepgram Nova-3) are indexed in pgvector or Elasticsearch. Students search the lecture by phrase or concept and jump to the exact timestamp. Chapter markers are auto-generated from topic-shift detection (slide changes, Q&A start, breakout-room transitions). Deep links work from anywhere — mobile, web, embedded in the LMS course page.
Proctoring is a separate classroom mode for assessments: webcam attention tracking, gaze direction, ambient audio anomaly detection, lockdown browser (custom or Respondus integration). Every event is logged to an audit-grade chain-of-custody record. FERPA / GDPR posture is enforced at every layer: encryption at rest and in transit, role-based access by LMS role, audit logs on every gradebook write, data residency pinned per region, retention windows configurable per course. We sign DPAs before any engagement.
Yes. SeamlessM4T or DeepL Voice runs as a parallel pipeline producing live translated audio tracks delivered over WebRTC. Captions are translated separately so students can read in language A while listening in language B. End-to-end translation latency is sub-second under typical conditions. Multilingual classes (international universities, language-immersion programs, global L&D) are a common pattern.
Yes. A new LiveKit / mediasoup classroom runtime comes up next to the existing Zoom integration, traffic is split by course or feature flag, the existing Zoom stays as fallback while QoE and gradebook posting are validated. Once metrics match or beat baseline, traffic shifts over.
Yes. K-12 mode includes COPPA-compliant defaults (no PII collection beyond what the district approves), parent / guardian visibility dashboards for elementary grades, AI-flagged disengagement so the teacher sees who's drifting in real time, optional lockdown browser for assessments. Districts configure retention, recording policy, and parent-facing reports per their own policy.
A virtual classroom in production, with LTI 1.3 launch, an xAPI / cmi5 back-channel, live captions and recording, starts from $58,000. Classrooms at scale, with multi-region clusters, a full AI pipeline and proctoring, start from $100,000.
You do. Models, training data, infrastructure code, classroom UI, recording archive, and AI overlays are all delivered to your repositories and your storage under your name. Fora Soft retains no claim on the IP or the data. The benefit of custom development over SaaS is exactly this: the recordings, the gradebook data, and the unit economics live on your balance sheet rather than the vendor's.
Three shapes: handover to your in-house team (most common on large platforms with EdTech engineers of their own); an ongoing SRE / AI-tuning retainer (typical when EdTech engineering isn't on your roadmap); or fixed-scope quarterly improvement cycles (new classroom types, new LMS integrations, AI feature upgrades). All three are scoped after the initial build, not bundled.
You'll get a realistic estimate, a technical recommendation, and an outline of next steps. No obligation. NDA before any access to your code, recordings, or LMS data.
Then we come back with next steps.
We ask the questions that change the budget.
Budget and architecture notes. Yours to keep.