Services Hire Developers Pricing About Blog Case Studies Book Free Consultation →
Industry: EdTech

Software development for edtech

Engineering for tutoring agencies, online schools, course platforms, and corporate learning teams. We ship custom LMS platforms with live video classrooms, AI graded assessments, proctoring, and certificate issuance, and we treat COPPA, GDPR-K, FERPA, and accessibility as design constraints rather than as a checklist tacked on the week before launch.

Get free assessment → View case studies

Industry challenges we solve

EdTech is a load problem, a trust problem, and a regulatory problem at the same time. The platform has to handle a peak that arrives the hour before a class starts, satisfy parents and regulators on data collection, and stay usable for learners across a wide range of devices and abilities.

⚠
Legacy LMS UX nobody wants to use. Moodle, Blackboard, and the older course platforms feel like 2010. Tutors switch to Zoom plus Google Docs, the data fragments, and the LMS becomes a glorified gradebook.
⚠
Live video that breaks at term start. The infrastructure handles fifty concurrent students in dev. The first week of term brings five thousand. Audio drops, video freezes, and the support inbox explodes by 9am Monday.
⚠
COPPA, GDPR-K, and the Children's Code pile up. A US, UK, and EU rollout means three overlapping regulatory regimes for under-13s and under-16s. Default settings, age estimation, and parental consent flows have to be different per jurisdiction.
⚠
Proctoring that crosses into surveillance. Heavy handed remote proctoring records continuous video of a learner's bedroom. Press coverage follows. Trust collapses faster than the platform can be redesigned.
⚠
Accessibility tacked on at the end. A WCAG audit lands two weeks before launch. Half the components fail screen reader testing. Procurement at universities and councils stalls until it is fixed.
⚠
Learner analytics that nobody trusts. Dashboards show pass rates and time on task but cannot answer "is this student about to drop out". The data is there, but the model has never been built.
Capabilities

Our edtech capabilities

🎓

Custom LMS & course authoring

Multi tenant LMS with course authoring, lesson sequencing, prerequisites, drip release, branching scenarios, SCORM 1.2 / 2004 import and export, xAPI statements into a Learning Record Store, and LTI 1.3 Advantage so you can plug into or host third party tools.

📹

Live video classrooms

Sub 200 ms small group sessions on LiveKit, large lecture broadcast on LL-HLS via Mux or Cloudflare Stream, real time chat on Socket.io, shared whiteboard with CRDT conflict resolution, screen sharing, and recording with retention policies that respect minor data law.

🎬

Video streaming pipelines

Upload, transcode (FFmpeg or Mux), HLS / DASH packaging with DRM (Widevine, FairPlay, PlayReady) where required, captioning and translation via Whisper plus an editorial pass, chaptering, and viewing analytics that connect back to learner progress.

🤖

AI grading & tutoring

Rubric driven grading of subjective answers with structured outputs, calibrated against a held out human-graded set, with confidence based routing to a human marker. AI tutoring chat constrained to the course corpus via RAG so it never goes off topic or hallucinates a textbook reference.

🛡️

Proctoring & integrity

Browser lockdown, tab switch detection, copy paste blocking, periodic webcam stills with on-device face presence checks, similarity detection across submissions, and integration with Honorlock, ProctorU, or Proctorio when full recorded proctoring is mandated.

📜

Certificates & learner analytics

Tamper evident PDF and verifiable digital credential issuance (Open Badges 3.0), per learner mastery dashboards visible to learners, tutors, and parents, attendance and progression analytics, at-risk learner scoring, and CSV / API exports for institutional reporting.

Compliance & regulatory considerations

EdTech regulation is unusually dense because the audience often includes minors and the data often crosses borders. We design to the strictest applicable regime per learner cohort rather than to the lowest common denominator.

COPPA (US under-13) FERPA (US institutions) GDPR-K (EU under-16) UK Age Appropriate Design Code UK GDPR + DPA 2018 CCPA / CPRA WCAG 2.2 AA SOC 2 Type 1 (in progress)

Children's data is the load bearing constraint. COPPA in the US requires verifiable parental consent before any personal information is collected from a child under 13, prohibits behavioural advertising, and forces a clear deletion path that flows through every sub-processor. GDPR-K in the EU sets the digital consent age between 13 and 16 depending on member state, with parental consent required below that age. The UK Age Appropriate Design Code (Children's Code) layers fifteen additional standards including high default privacy, no nudge patterns, age estimation where appropriate, and minimised data sharing. We treat child accounts as a different schema from adult accounts so the wrong control can never apply to the wrong record, and we maintain a per-jurisdiction matrix of defaults so a US, UK, and EU rollout does not require three different code paths.

FERPA matters the moment you sell into US K-12 districts or universities. We build to the FERPA Direct Service Agreement model: the institution remains the data controller, we are the processor under instruction, the school official exception is documented, and parental access to records is delivered through a dedicated portal rather than through customer support. Auditable access logs survive a Department of Education enquiry.

Accessibility is procurement-blocking, not optional. Universities, councils, and most regulated employers will not buy a platform that fails WCAG 2.2 AA. We design to AA from day one and verify with both automated tooling (axe-core, Lighthouse, Pa11y) and manual screen reader testing on NVDA and VoiceOver. Captioning, keyboard navigation, focus management on modal dialogs, sufficient colour contrast, and reduced motion preferences are part of the component library, not a remediation phase.

Tech stack we use for edtech

Built for video, content, and a peak that arrives the hour before class. We default to managed video infrastructure where it makes commercial sense and self host where unit economics demand it.

Node.js 20 + TypeScript React 18 + Next.js React Native LiveKit (WebRTC SFU) Mux / Cloudflare Stream FFmpeg + HLS / DASH PostgreSQL 16 + row level security Redis + BullMQ OpenAI + Anthropic Claude pgvector for RAG tutoring SCORM / xAPI / LTI 1.3 Stripe Connect for agency billing
New Product Development UK EdTech startup · 10 weeks · 4 engineers

EdTech LMS platform built from scratch in 10 weeks for the January term

A London EdTech startup needed a multi tenant LMS for GCSE and A Level tutoring with live video, shared whiteboards, AI grading, and 15,000 concurrent students before term began. We architected on LiveKit and PostgreSQL row level security, delivered Stripe Connect billing for the agency network, and load tested at 15K concurrent connections before the first agency sent a single student. Phase 1 shipped in 10 weeks with a 4 person team using Claude Code and Cursor throughout.

10 wksConcept to production
15KConcurrent users supported
<200msVideo latency (P95)
92%AI grading accuracy

Read full case study →

FAQ

Common questions

Yes. For US under-13 audiences we treat COPPA as the design constraint, not an afterthought: verifiable parental consent flows (email-plus method or government ID where required), no behavioural advertising, minimised data collection, and a clear deletion path that flows through to every sub-processor. For EU and UK under-16 audiences we apply GDPR-K and the UK Age Appropriate Design Code (Children's Code) standards, including age estimation, default high privacy, and no nudge patterns. The data model separates child accounts from adult accounts at the schema level so the wrong control can never be applied to the wrong record.
For interactive small-group sessions (2 to 50 participants) we use LiveKit on a self-hosted SFU or LiveKit Cloud, with sub 200 ms latency, simulcast, and adaptive bitrate. For one-to-many live lectures (hundreds to tens of thousands of viewers) we shift to HLS or LL-HLS via Mux or Cloudflare Stream, with a real-time chat sidecar on Socket.io. Recordings go to S3 with PHI/PII aware retention. We load test with k6 and Artillery against the actual SFU before term start, not after.
Yes. SCORM 1.2 and 2004 packages can be authored, imported, and tracked in the LMS. xAPI (Tin Can) statements flow into a Learning Record Store either built in PostgreSQL or integrated with a third party LRS. LTI 1.3 with LTI Advantage (Names and Role Provisioning, Assignment and Grade Services, Deep Linking) lets your platform act as a tool inside Canvas, Moodle, Blackboard, or Schoology, or host third party tools inside your own LMS.
Yes, and we are deliberate about the trade-off. Lightweight proctoring (browser lockdown, tab-switch detection, copy-paste blocking, periodic webcam stills with on-device face presence detection) covers most academic integrity use cases without recording continuous video of a learner's bedroom. Where a regulator (professional certification body, university exam) requires recorded proctoring, we use Honorlock, ProctorU, or Proctorio integrations rather than building our own. Either way, the candidate sees exactly what is being recorded and can choose an alternative assessment if available.
The tutor or course author defines a rubric (criteria, weightings, exemplar answers at each grade band). An LLM scores each free text answer against the rubric, returns a structured score per criterion plus a written justification, and flags low confidence cases for human review. We calibrate against a held out set of human-graded answers per assessment, publish the agreement rate (typically 88 to 94% for well written rubrics), and never let AI grades go out without human sign-off on borderline cases. Hallucination is bounded by structured outputs and refusal patterns when the rubric does not fit the answer.
Yes, and the deadline drives the architecture. We did exactly this for a UK EdTech: 10 weeks from architecture sign off to a multi tenant LMS supporting 15,000 concurrent students with live video, AI grading, parent dashboards, and Stripe Connect billing for the agency network. The trick is to lock scope hard at week one, run weekly load tests from week three, and keep AI agents focused on scaffolding and tests so the senior engineers can spend their time on the parts that actually break under load.
Free consultation, no commitment

Ready to ship?

Tell us about your project. Written scope, timeline and cost estimate within 48 hours.

Book a free consultation → Explore services →
Chat with us on WhatsApp