AI course generator that adapts to the learner's level

The project began as a solo experiment to solve a clear limitation: tools like ChatGPT can generate educational content, but they can't structure learning in a way that is progressive, personalized and pedagogically consistent. The idea was to build a system that generates complete courses inside LearnDash, based on the user's real level and the skills needed to reach a specific goal.

Problem:

• Users don't know how to structure their own learning (what to study, in what order, at what depth)
• AI tools generate isolated content but don't build learning journeys
• Traditional courses are generic and don't adapt to the learner's real level
• There is no link between main skills and the adjacent skills they require
• There is still no truly educational AI content generation tool

Solution:

• We built a system that generates personalized courses with AI
• The user defines a focus skill (for example, decision-making)
• The AI generates a complete course (up to 20 lessons) with a structured pedagogical progression
• The course is created automatically in LearnDash, ready to use

Technologies used:

• WordPress + LearnDash
• Custom plugin
• AI-agnostic integration: any AI can be plugged in Course creation screen: say which skill you want to learn

A skill-driven architecture (not a content-driven one)

The main structural problem was keeping the system from becoming just a "text generator". The solution was to model the system with skills as the central unit, not courses.

• Course = container
• Lesson = reusable asset
• Skill = the main semantic unit The architecture works like a knowledge graph:
• Skills have levels (rubrics)
• Skills connect through dependencies (skill_edges)
• The system identifies gaps between the current and desired level
• The course is generated to fill those gaps Critical decision: avoid duplicated content by using content_hash and reuse existing lessons whenever possible. There is also a very strong layer of instructional design guiding how content is created. It's not enough to ask the AI to generate content: you have to tell it how. This lowers cost, keeps things consistent and allows scale. Secondary skills: the system probes your knowledge

Development and generation journey

The system was designed as a progressive flow (almost a "smart wizard"):

1. The user describes what they want to learn
2. The AI extracts the main skill
3. The AI generates the level rubric
4. The user sets their current level and goal
5. The AI identifies adjacent skills
6. The user assesses themselves on those skills
7. The AI generates a complete roadmap
8. The system turns the roadmap into a course in LearnDash Course generation happens in a modular, asynchronous way:

• One endpoint creates the base course
• Each lesson is created separately (to avoid timeouts)
• Content is generated as JSON and then materialized in the database

Features built around real learning

Instead of focusing only on content, the system was designed around pedagogical principles: 1. Deep personalization

• A course based on the user's real level
• Considers multiple skills, not just the main one 2. A mandatory pedagogical structure Every lesson follows defined patterns:
• A clear objective
• Difficulty level
• Estimated time
• Skill context 3. Balanced lesson types
• Learn (concept)
• Practice (execution)
• Reflect (internalization)
• Interact (external feedback) Mix:
• 30% learn
• 35% practice
• 20% reflection
• 15% interaction 4. An integrated quiz system
• Knowledge checks spread throughout the course
• Automatic creation via WPProQuiz
• Direct linking to lessons 5. Full generation inside LearnDash The system doesn't just generate content. It creates the whole structure:
• Course (sfwd-courses)
• Lessons (sfwd-lessons)
• Topics (sfwd-topic)
• Quizzes and questions All through an automated pipeline, with no manual intervention

Results

• Complete courses created automatically in minutes
• A sharp drop in the need for instructional designers for the initial structure
• A scalable system built on content reuse
• More efficient learning thanks to adaptation to the user's real level Lessons learned:
• Real personalization requires knowledge modeling, not just generative AI
• Without structural control, AI produces redundancy and low quality
• Architecture (skills + graph) matters more than the content itself This whole model also allows a company to insert its own ready-made content and define rules.