A production-ready exam preparation and course management platform built on Next.js 15, Payload CMS v3, and PostgreSQL. The platform covers the full lifecycle from content creation to student exam practice - with a rich admin panel, an adaptive learning engine, Anki-style spaced repetition flashcards, and a detailed audit log.
Technical reference: all detailed setup, imports, testing, architecture, and stack docs live under LearningPlatform/documentation/ (start with the index file linked here).
| Area | Highlights |
|---|---|
| Content management | Subjects → Courses → Modules → Lessons → Tasks; draft/publish (isPublished) applies to courses, modules, lessons, and tasks (subjects are taxonomy only) |
| Lesson builder | 6 composable content blocks: Rich Text, Image, Math (KaTeX), Callout, Video, Table |
| Task engine | Multiple Choice, True/False, Open-Ended with optional auto-grading |
| Adaptive learning | Tag-based weakness scoring surfaces personalized practice tasks |
| Spaced repetition | Full SM-2 algorithm - NEW / LEARNING / REVIEW / RELEARNING / MASTERED states |
| Audit trail | Persistent activity log for every admin action, viewable and filterable in the panel |
| Media | S3-compatible object storage with signed URL delivery and usage tracking |
| Security | JWT revocation, rate limiting, per-request CSP nonces, Zod validation throughout |
| Progress tracking | Per-user lesson completion and task grading stored at every granularity level |
Administrators manage the platform through a dedicated admin panel at /admin, built with Payload CMS and a custom Next.js UI layer.
The sidebar (see LearningPlatform/components/admin/sidebar.tsx) includes:
- Dashboard — admin home
- Lessons — block-based lesson authoring (see Lesson Builder); courses and modules are edited inside these flows, not as separate sidebar entries
- Subjects — top-level taxonomy
- Tags — canonical tags for tasks and flashcards (recommendations)
- Tasks — assessments with prompts, solutions, points, and tags
- Flashcards — SRS cards (deck organization is supported in the product; see
/api/flashcard-decks) - Media — uploads and library management
- AI Agent — experimental course generation workspace
- Logs — audit trail (see Activity Log)
- Users — user list and Pro entitlement (
isPro) - Settings — admin UI preferences (theme, reading size)
Subjects do not have an isPublished field; they are always available to admins for organization. Courses, modules, lessons, and tasks each have their own isPublished flag. Students only see content that is published along the path they are viewing.
Subject ──► Course ──► Module ──► Lesson ──► Task
(published) (published) (published) (published)
- A Course can be hidden in draft while its module structure is being built.
- A Lesson can be marked published independently - useful when releasing content progressively.
- A Task remains invisible to students until individually published.
- Admins always see all content regardless of publish state.
This gives content authors full control over what students see and when, without deleting content or creating branches.
Lessons are assembled from a sequence of typed content blocks. Admins add, reorder, and remove blocks freely in the Payload admin editor. Six block types are available:
Rich text powered by the Lexical editor. Supports headings, bold/italic, bullet and numbered lists, inline code, and links.
An image from the media library with configurable width (Small 400 px / Medium 600 px / Large 800 px / Full) and alignment (Left / Center / Right), plus an optional caption. Images are served through /api/media/serve/:filename (local file response or redirect to a signed S3 URL when cloud storage is configured).
A LaTeX formula rendered with KaTeX on the client. Supports both display mode (centred block equation) and inline mode for formulas embedded in surrounding text. An optional descriptive note is displayed beneath the formula.
Example: x = \frac{-b \pm \sqrt{b^2 - 4ac}}{2a}
A highlighted box in one of three semantic variants:
info- blue, for supplementary informationwarning- amber, for caveats and important notestip- green, for helpful suggestions
Each callout contains its own rich text body.
An embedded YouTube video with selectable aspect ratio (16:9 widescreen or 4:3 standard), an optional title displayed above the player, and an optional caption below.
A structured data table with optional column headers. Supports a caption displayed above the table. Table data (headers and rows) is stored as JSON and rendered as a native HTML table.
Each lesson can have any number of associated tasks. Three task types are supported:
| Type | Description | Grading |
|---|---|---|
| Multiple Choice | A list of labelled options; one is correct | Automatic - matched against correctAnswer |
| True / False | Binary choice question | Automatic - matched against "true" or "false" |
| Open Ended | Free-text answer | Manual by default; optional auto-grade mode normalises and compares text case- and punctuation-insensitively |
Every task can include:
- A rich text prompt with full formatting
- An optional question image from the media library
- A solution / explanation in rich text, shown after submission
- An optional solution image and/or solution video URL (YouTube)
- Tags linking the task to the knowledge taxonomy
- A configurable point value and display order
The recommendation engine builds a per-user skill model from task submission history:
- Tag weakness scores - for each tag the student has encountered, the engine computes a score based on recent answer accuracy and recency weighting.
- Personalized task queue - tasks tagged with the student's weakest knowledge areas are surfaced first at
/api/recommend/tasks. - Tag stats cache - per-user tag statistics power recommendations and practice (
/api/practice/session); results are cached (short TTL) and invalidated on new task submissions.
The full algorithm is documented in ADAPTIVE_LEARNING.md.
The flashcard system implements the SM-2 algorithm with Anki's state-machine extensions.
Each card tracks an independent per-user state:
| State | Description |
|---|---|
NEW |
Never studied |
LEARNING |
In short learning steps (minutes-based scheduling) |
REVIEW |
Long-term spaced repetition (days-based scheduling) |
RELEARNING |
Failed a REVIEW card; returns to short steps before re-entering REVIEW |
MASTERED |
Interval ≥ 21 days (continues to be reviewed normally) |
When reviewing a card the student chooses one of four answers - Again, Hard, Good, Easy - and the algorithm updates the interval and ease factor accordingly. Learning steps are configurable (default: 1 min → 10 min → graduate to REVIEW at 1 day). Ease is clamped at a minimum of 1.3 to prevent cards from becoming impossibly infrequent.
Cards are linked to tags, so the SRS session can be filtered to a particular topic.
A persistent activity_logs table records every significant admin action. The admin Logs page at /admin/logs displays a paginated, filterable table of all entries.
Actions covered include:
- Authentication -
USER_REGISTERED,USER_LOGIN,USER_LOGOUT,USER_LOGIN_FAILED - Subject / Course / Module / Lesson / Task - created, updated, published, unpublished, deleted
- Flashcard & Tag - created, updated, deleted
- Media - uploaded, deleted
- Pro / assistant (where applicable) - lesson assistant usage and admin Pro flag updates
Each log entry stores the actor's user ID and email, the affected resource type and ID, a UTC timestamp, and an optional JSON metadata payload (e.g. the resource title).
Filters available in the admin panel: action type, actor user ID, date range. All filters are reflected in the URL for bookmarkable views.
The logging design is fire-and-forget - a logging failure never disrupts the primary operation.
Full reference: LOGGING_SYSTEM.md
All uploaded files are stored in an S3-compatible object store when configured (AWS S3, Cloudflare R2, Railway Object Storage, or any compatible API); otherwise media can be stored locally under public/media.
- Admins upload via
POST /api/media/upload; assets are tracked in the Payload Media collection. - Delivery goes through
/api/media/serve/:filename(signed URL redirect to S3 when enabled, or local file serve). - Usage is tracked per file so admins can see which lessons and tasks reference a given asset before deletion.
- Used by lesson image blocks, task question/solution media, and related fields.
Canonical index (same structure as the repo): LearningPlatform/documentation/README.md
PLATFORM_FEATURES.md— Admin capabilities, draft/publish workflow, lesson blocks, task types, media behavior.ADAPTIVE_LEARNING.md— Recommendations, weak-tag analytics, practice sessions, spaced repetition.AI_COURSE_GENERATION.md— Admin AI Agent flow, providers, testing notes, cost guidance.
LOCAL_DEVELOPMENT.md— Prerequisites, full local setup, quick start, Docker, local URLs.CONTENT_IMPORTS.md— Bulk imports: data shapes, npm scripts, troubleshooting, Docker.TESTING.md— Vitest and Playwright commands, suite layout, env flags, coverage baseline.
DATABASE_ARCHITECTURE.md— Dual-schema design, models, indexing, integrity.SECURITY_ARCHITECTURE.md— Auth, authorization, rate limiting, CSP, sessions.LOGGING_SYSTEM.md— Activity log actions, schema, admin usage.TECHNOLOGY_STACK.md— Framework, data, auth, testing, deployment layers.
prompts/creation_prompt.MD— Large template for LLM-generated import-shaped content.screenshots/— PNG assets (for example referenced fromAI_COURSE_GENERATION.md).