Learning Management on Your Own Infrastructure
Three LMS platforms for demos — adaptive language courses, structured learning paths, internal knowledge bases. Built on open-source foundations, deployed on Edge.
Live DemoThree LMS platforms for demos — adaptive language courses, structured learning paths, internal knowledge bases. Built on open-source foundations, deployed on Edge.
The Foundation
Knowledge Core
Open Source TemplateThe shared foundation of all LMS instances: a pnpm monorepo with docs and course apps, shared components and a type-safe content model. Clone, customise, deploy — without vendor lock-in.
- pnpm Workspaces — Docs + Courses in one repo
- Type-safe content model with Zod schema
- Shared Component Library
- Multi-subdomain deployment
LMS as Live Demo
Three instances built on Knowledge Core — different content, the same scalable foundation.
Language Toolbox
Adaptive Language Course LMSLanguage courses with an adaptive learning model: the platform adjusts difficulty and repetition intervals individually — following a three-phase cycle of comprehensible input, active production, and corrective feedback. No fixed course plan, but a learning logic that adapts to each user's progress.
- Adaptive spaced repetition algorithm
- Grammar rule sets per language
- Interactive quizzes and exercises
- Persistence via Edge DB
- DE · EN · ES · FR · IT
- Mistral AI integration (Roadmap Phase 5)
Learning Space
General Learning PlatformStructured learning paths across multiple subject areas — with cross-device progress sync and MDX-based content structure. 628 lessons, 13 paths, one repository.
- Learning paths with defined dependencies
- Cross-device sync via KV store
- Progress display per course and path
- MDX lessons with quizzes and exercises
- KaTeX for mathematical formulae
- Multiple subject areas in parallel
Upscale
Modular Knowledge PlatformStructured learning paths at secondary-school level — six subject areas, interactive lessons and KaTeX formulae for mathematics and science content. Demonstrates how Knowledge Core can be adapted for subject-specific depth.
- Six subject areas (Maths, Physics, Chemistry etc.)
- KaTeX for mathematical formulae
- Structured learning paths per subject
- Interactive quizzes and exercises
LMS Core Features
What a complete learning platform needs — and what is implemented here.
Course & Lesson Structure
Hierarchical content model: Courses → Sections → Lessons. MDX files as single source of truth, type-safe via Zod.
Learning Paths
Pre-structured sequences across multiple courses with defined dependencies — guided entry, visible progress.
Quiz Engine
Multiple choice, true/false, free text — with immediate feedback and progress tracking per lesson.
Progress & Sync
Learning progress synchronised across devices via KV store — no separate user backend required.
Protected Areas
Areas can be secured via middleware — e.g. via Cloudflare Access, a custom auth layer or SSO. Adaptable to any infrastructure.
Multi-Subdomain
Docs, courses and portals under different subdomains — from one repository, one deployment pipeline.
Didactic Foundation
Content alone does not teach. To retain something, learners must actively work with it — retrieve, apply, repeat. That is why the learning architecture follows principles from cognitive science rather than simply providing content.
Learning Cycle
Each learning unit follows the three-step cycle established in second language acquisition research: contextual input, active production, adaptive feedback.
The grammar rules driving this cycle are modelled in machine-readable form using the CUE configuration language — validatable, versioned, extensible.
Spaced Repetition
Repetition intervals adapt to individual learning progress. Difficult content appears more frequently; mastered content less so.
- Adaptive difficulty curves per learner
- Persistent learning history in Edge DB
- No fixed course plan — pace determines the path
- Grammar rule sets as machine-readable data model — validated with the CUE configuration language
Active Recall Instead of Passive Reading
All three platforms are structured so that learners actively engage with content. Every lesson ends with a knowledge check. Learning paths build on each other and make dependencies explicit — scaffolding, not self-study without orientation.
Technology Stack
Deployed at the Edge — no central server, low latency, simple deployment. Cloudflare is the reference implementation; the pattern works on any Edge platform.
Frontend & Content
- Astro / Node.js — SSR + Static, Island Architecture; runs on Node hosting or Edge
- MDX + Metadata as Static — course content scales without a database; thousands of lessons stay performant
- Database integration — optional; PostgreSQL, MySQL, SQLite or REST API can be integrated without difficulty
- TypeScript + Zod — type-safe content models
- pnpm Workspaces — monorepo with shared packages
Edge Deployment & Persistence
Runs at the Edge — no dedicated servers, close to the user. Reference implementation on Cloudflare; the pattern is transferable to other Edge platforms.
- Edge Runtime — e.g. Cloudflare Workers, Vercel Edge, Deno Deploy
- SQLite at the Edge — e.g. Cloudflare D1, Turso, PlanetScale
- Key-Value Store — e.g. Cloudflare KV, Vercel KV, Upstash
- AI Inference — e.g. Workers AI, OpenAI, local model (Phase 5)
Deployment Architecture
docs.company.com Documentation, API reference
learn.company.com Courses, learning paths, quizzes
intern.company.com Protected corporate area
All from one repository — Vercel, Netlify or Cloudflare Workers
Development Roadmap
Where the platforms stand today — and where they are heading.
Core LMS Engine
MDX-based content model, course and lesson structure, quiz engine with multiple choice and free text.
Adaptive Spaced Repetition
Input–Output–Feedback learning cycle for language courses, cross-device progress persistence via KV store and Edge DB.
Multi-Language & Learning Paths
5 target languages (DE, EN, ES, FR, IT), 13 learning paths, 628+ lessons live on learn.casoon.dev.
Edge Persistence
Progress is stored anonymously — on first launch you receive an ID that lets you restore your state across devices. No registration, no personal data.
AI-Assisted Language Adaptation
Mistral AI via Workers AI — dynamically generated exercises, personalised feedback and adaptive difficulty curves without manual content authoring.
Target vision Phase 5
AI Layer: Cloudflare Workers AI + Mistral
The grammar rules and course data are already machine-readable — Mistral can build directly on top of them. All inference runs via Cloudflare Workers AI at the edge: no external API round-trip, no added latency, no data leaving the system.
Mistral generates new tasks directly from the existing CUE grammar rules — fill-in-the-blanks, transformations, translations without manual authoring.
Instead of predefined error messages: individual explanations tailored to the specific mistake and the learner's level.
Dialogue exercises in the target language — Mistral leads the conversation, detects errors and corrects them in the learning rhythm.
New content is classified by difficulty and slotted into the spaced repetition schedule — without manual annotation.
Context
Conceptual work with working prototypes
What is described here is not a finished product. The platforms run, the learning logic works — but there are still bugs, open questions and areas that need further development.
Cloudflare is the current technical foundation — short deployment cycles, edge infrastructure without dedicated servers, D1 and KV as a straightforward persistence layer. Technically there is no lock-in: the architecture can be moved to European hosting providers that offer the same primitives with manageable effort. A GDPR assessment is a standard requirement for any production deployment regardless of which platform is ultimately chosen.
The underlying goal is clearer than today's implementation: a learning system that requires no licensed components. No recurring SaaS costs, no proprietary dependencies — just labour, low hosting costs, and Mistral as a privacy-friendly, cost-efficient language model layer.
Mistral is not just a technical choice but a principled one: a European model suited for GDPR-compliant deployments — whether self-hosted or via providers with EU data centres.
Build an LMS Platform
Knowledge Core is available as an open-source template — or as a tailored platform with your own branding and content.
Clone the Template
Ready to use immediately. Open source on GitHub.
git clone https://github.com/casoon/knowledge-core.git
pnpm install
pnpm devCustom Implementation
A tailored LMS with your own content, branding, access control and deployment infrastructure.
Enquire about a project