The Challenge
A London-based EdTech startup had secured seed funding to build an online tutoring platform for GCSE and A-Level students across the UK. Their vision was a modern alternative to legacy platforms like Google Classroom and Moodle, purpose-built for one-to-one and small-group live tutoring with meaningful AI assistance.
No existing product fit their model. Off-the-shelf LMS tools were designed for university-scale lecture recording, not interactive tutoring sessions. The founders needed live video with real-time whiteboard collaboration, instant assessment feedback, and student progress dashboards that parents and tutors could share.
Tight launch window. The platform needed to be live before the January term to onboard tutoring agencies already committed to piloting it. That gave the team roughly 10 weeks from architecture sign-off to production launch, with no room for a phased rollout.
Scale from day one. Three tutoring networks had agreed to pilot, bringing an estimated 15,000 concurrent students during peak evening hours. The infrastructure had to handle real-time video, WebSocket events, and AI inference without degrading under load.
The Approach
RG INSYS assembled a 4-person AI-native team: one architect, two full-stack engineers, and one DevOps engineer. AI coding agents (Claude Code, Cursor) handled scaffolding, test generation, and boilerplate throughout the project.
Multi-tenant architecture: Each tutoring agency operated as an isolated tenant with their own branding, user management, and billing. We used a shared database with row-level security in PostgreSQL, keeping infrastructure costs low while ensuring data isolation.
Live video via WebRTC: We integrated a WebRTC-based video layer using LiveKit for low-latency group video. A shared digital whiteboard (built on Canvas API with CRDT-based conflict resolution) allowed tutors and students to annotate simultaneously during sessions.
AI-graded assessments: Tutors could create quizzes mixing multiple-choice, short-answer, and long-form questions. For subjective answers, we integrated an LLM grading pipeline that scored responses against rubrics defined by the tutor, flagging edge cases for human review.
Timeline: Week by Week
Weeks 1–2: Architecture design, database schema, CI/CD pipeline, authentication (SSO + magic link). Multi-tenant data model with row-level security. Component library setup in React.
Weeks 3–4: Core platform: course management, lesson scheduling, student enrollment, tutor dashboards. WebRTC video integration with LiveKit. Real-time whiteboard MVP.
Weeks 5–7: Assessment engine with AI grading. Student progress analytics with parent-facing dashboards. Payment integration (Stripe) for agency billing. Notification system (email + in-app).
Weeks 8–10: Load testing at 15K concurrent connections. Security audit. Beta rollout with two pilot agencies. Bug fixes, performance tuning, and production deployment.
Tech Stack
- Backend: Node.js 20, Express, TypeScript, Prisma ORM
- Frontend: React 18, TypeScript, Tailwind CSS, Zustand
- Real-time: LiveKit (WebRTC), Socket.io, Redis Pub/Sub
- AI: OpenAI API for assessment grading, LangChain
- Database: PostgreSQL 16, Redis 7
- Infrastructure: AWS (ECS Fargate, RDS, ElastiCache, CloudFront, S3)
- Payments: Stripe Connect for multi-tenant billing
- AI tooling: Claude Code, Cursor IDE
Results
Key Features Delivered
- Live video classrooms: Low-latency group video with screen sharing, recording, and a real-time collaborative whiteboard. Sessions automatically archived for student review.
- AI-graded assessments: Tutors define rubrics; the LLM scores subjective answers and provides written feedback. Edge cases flagged for manual review, reducing grading time by 75%.
- Student progress dashboards: Real-time analytics visible to students, tutors, and parents. Topic mastery tracking, attendance records, and personalised improvement suggestions.
- Multi-tenant billing: Stripe Connect handles per-agency billing with configurable pricing tiers, automated invoicing, and revenue sharing splits.
Building an EdTech product?
We ship production-ready platforms in weeks, not months. Get a scope, timeline, and cost estimate within 48 hours.
Book Free Consultation →