sherab-custom-plugin

Custom Open edX plugins for the Sherab / WeBuddhist Academy platform, packaged together as a single installable distribution (custom-extensions). The package bundles four independent Django apps that plug into the LMS and/or CMS via the Open edX plugin framework (edx_django_utils.plugins) — no edx-platform core code is modified.

App	Runs in	Purpose
ai_course_creator	CMS (Studio)	Conversational AI ("Sherab") that designs and generates whole courses, and edits existing sections, using Google Gemini.
course_partnerships	LMS	Partner organizations, centers, categories, and course-creator profiles, with branding assets and a mobile API.
user_extension	LMS	Extends user profiles with partner relationships; auto-approves ID verification on registration.
wishlist	LMS	Personal per-user course wishlist.

---

Installation

The plugin is installed into the edx-platform virtualenv. Under Tutor it is mounted and listed as an extra pip requirement so it is picked up on image build / dev mount.

config.yml:

OPENEDX_EXTRA_PIP_REQUIREMENTS:
  - google-api-python-client
  - google-genai
MOUNTS:
  - /path/to/sherab-custom-plugin   # dev: live-mounted into LMS & CMS

Python dependencies are listed in requirements/common.in and installed automatically by setup.py.

The apps register through the entry points in setup.py:

cms.djangoapp:  ai_course_creator, course_partnerships, user_extension
lms.djangoapp:  course_partnerships, user_extension, wishlist

After installing or changing models, run migrations for the relevant process, e.g.:

tutor dev exec cms ./manage.py cms migrate ai_course_creator
tutor dev exec lms ./manage.py lms migrate course_partnerships

---

ai_course_creator

A Studio-only (CMS) conversational assistant — "Sherab" — that helps a course creator design a course through chat, then generates the real Open edX course structure for them. It also powers a per-section "Edit with SherabAI" sidebar for improving an existing section. All LLM calls are server-side only; the Gemini API key is never sent to the browser.

Two flows

1. Whole-course creator. From an empty course outline the creator chats with Sherab through a guided, four-phase flow (Learner → Transformation → Materials & Assessment → Generate). When the conversation is ready, Sherab generates the full outline and writes it into the course as draft sections, subsections, units, and components.

2. Per-section editor. From a populated outline the creator opens a section in the "Edit with SherabAI" sidebar. The conversation is anchored on that section's zero-to-hero transformation (what a learner can do after the section that they couldn't before). Sherab proposes concrete edits; nothing is written until the creator clicks Apply changes, and edits are strictly scoped to that one section.

Components

• models.py

  • ChatSession — one conversation, unique per (user, course_id, section_locator). section_locator is empty for the creator flow and the chapter usage key for the per-section editor, so each section keeps its own thread. Caches the generated course_json, the display current_phase, the generation_status lifecycle, and the created_section_locators (for rollback / detecting existing Sherab content).
  • ChatMessage — a user/assistant turn within a session.
  • UploadedMaterial — course material shared by the creator (file/link/text), stored as extracted plain text (the binary is not retained).

• services/llm_client.py — thin wrapper over the google-genai SDK. Streams replies (stream_reply), makes JSON-mode generation calls (complete_json), loads the bundled skill as the system instruction, and parses/strips the embedded COURSE_JSON and SECTION_EDITS marker blocks. Handles rate-limit (429), 503 "high demand", and oversized-request (413) errors with backoff/retry and user-friendly messages.

• services/generator.py — orchestrates two-stage, resumable course generation: a small skeleton call (titles + objectives + component plan), then one content call per section (lesson HTML + assessments). Yields progress events consumed by GenerateCourseView.

• services/course_builder.py — turns the generated COURSE_JSON into real Open edX structure by calling contentstore/modulestore directly (chapter → sequential → vertical → component). Builds CAPA problem OLX for several question types. Also exposes user_can_author (the author-permission check) and delete_sections (rollback).

• services/section_editor.py — read_section serializes one chapter subtree to a JSON tree (usage keys + content) for the LLM; apply_section_edits applies an operation list (rename / editContent / add / delete / reorder) strictly scoped to that chapter. Every usage key the model supplies is validated against the chapter's own descendant set before anything is mutated.

• services/materials.py — extracts plain text from PDF / DOCX / PPTX / plain text uploads, pasted text, and public URLs; truncates to the per-request context budget.

• skill/ and skill_section_editor/ — the SKILL.md prompt files (plus skill/references/*.md) that define Sherab's persona and conversation rules for the creator flow and the section editor respectively. Loaded as the model's system instruction.

• helpers.py — view helpers: SSE framing (sse), session lookup (get_or_create_session), edit-and-resend truncation (truncate_for_edit), the shared course-key + author-permission guard (require_course_author), and feature/ limit settings accessors.

API endpoints

All require an authenticated user; course-mutating endpoints additionally check author access. Mounted at the site root, so paths are absolute on the Studio host.

Method	Path	Purpose
POST	/api/ai-course-creator/chat/	Stream the assistant's next reply (SSE).
POST	/api/ai-course-creator/upload/	Add a material (file / link / text).
DELETE	/api/ai-course-creator/material/<pk>/	Delete an uploaded material.
POST	/api/ai-course-creator/generate/	Generate + write the course (SSE).
GET	/api/ai-course-creator/config/	Feature-flag state for the Studio UI.
GET / DELETE	/api/ai-course-creator/session/	Fetch (resume) or reset a conversation.
GET	/api/ai-course-creator/section-content/	Read one section's content tree.
POST	/api/ai-course-creator/section-chat/	Stream a per-section editor reply (SSE).
POST	/api/ai-course-creator/apply-section/	Apply the latest proposed section edits.

Settings

Defaults are set in settings/common.py / settings/production.py; the Gemini credentials are normally injected into CMS settings from Tutor config.yml via the configuration_plugin patch.

Setting	Default	Purpose
GEMINI_API_KEY	""	Google Gemini API key (server-side only).
GEMINI_MODEL	gemini-1.5-pro	Model id (e.g. gemini-2.5-flash-lite).
AI_COURSE_CREATOR_ENABLED	True	Master on/off switch for the feature.
AI_COURSE_CREATOR_MAX_UPLOAD_BYTES	25 MB	Max uploaded-material size.
AI_COURSE_CREATOR_MAX_CONTEXT_CHARS	16000	Max material text per request.

Tutor wiring — set the values in config.yml:

GEMINI_API_KEY: "AQ.xxxxxxxx"
GEMINI_MODEL: gemini-2.5-flash-lite

and inject them into CMS settings in configuration_plugins.yml:

patches:
  openedx-cms-production-settings: |
    GEMINI_API_KEY = "{{ GEMINI_API_KEY }}"
    GEMINI_MODEL = "{{ GEMINI_MODEL }}"

then tutor config save and restart CMS.

Note: the bundled skill is read once and cached (lru_cache) for the process lifetime, so restart the CMS after editing any SKILL.md to pick up changes.

The matching frontend lives in the frontend-app-authoring MFE under src/ai-course-creator/ (the creator modal and the section-editor sidebar).

---

course_partnerships

LMS-only. Manages partner organizations, their sub-centers, course categories, and instructor/course-creator profiles — with branding assets — and links them to courses. Provides partner/center detail pages and a mobile-app JSON API.

Models

• Partner — schools / partner organizations (logo, banner, rich-text description; can activate school-admin features).
• Center — sub-entities under a Partner (logo, banner, description).
• Category — course categories, optionally tied to a partner; can be shown on the homepage.
• EnhancedCourse — links a CourseOverview to its Partner / Center / Category.
• PartnerOrganizationMapping — maps a Partner to an Organization, with a mobile-app visibility toggle and optional display-name override.
• CourseCreator — instructor profiles (name, title, years of experience, bio, picture).

API endpoints

Method	Path	Purpose
GET	/schools/<slug>/	Partner detail page (centers, categories, courses, creators).
GET	/schools/<partner_slug>/<center_slug>/	Center detail page.
GET	/api/partners/	Mobile-app JSON of partner-organization mappings.

Management commands

Command	Purpose
assign_course_partners	Auto-assign partners to courses lacking one, from org→partner mappings.
check_partner_logos	Show current storage locations / URLs for logos and creator profiles.
check_storage_settings	Validate the S3 storage backend config for partner/center/creator assets.
fix_logo_paths	Correct malformed logo/banner paths in the DB (--dry-run supported).

Behaviour & settings

• Signals: on course publish, creates/updates the EnhancedCourse row and auto-assigns a partner from the org mapping; on course delete, removes it.
• settings/common.py adds ckeditor to INSTALLED_APPS for the rich-text fields.
• Asset storage backends are read from Django config (PARTNER_LOGO_BACKEND, CENTER_LOGO_BACKEND, COURSE_CREATOR_STORAGE_BACKEND), wired via the tutor-contrib-s3 plugin.

---

user_extension

LMS-only. Extends user profiles with partner-organization relationships and streamlines onboarding.

• Model: ExtendedUserProfile — OneToOne with User; associates users with a Partner and stores admin relationships with timestamps.
• Signals:
  • sync_extended_profile — create an ExtendedUserProfile whenever a User is created.
  • auto_approve_id_verification_on_registration — create an approved ManualVerification for every new user, bypassing the standard ID-verification flow.
• Management command: approve_id_verification — bulk-approve ID verification (--username, --all, --batch-size, --dry-run).
• No REST endpoints; no settings injected.

---

wishlist

LMS-only. A personal per-user course wishlist.

• Model: Wishlist — one entry per (user, course) pair.
• API endpoints:

Method	Path	Purpose
POST	/wishlist/change-status/	Add/remove a course (wishlist_action = add / remove).
GET	/wishlist/	List the authenticated user's wishlisted courses.

• No settings injected; no active signals.

---

Development

• Formatting: Black, line length 120 (see pyproject.toml).

• Making migrations (from the edx-platform dir / inside the container):

  ./manage.py cms makemigrations ai_course_creator
  ./manage.py cms migrate ai_course_creator

• Dev mount: with the plugin in MOUNTS, code changes are live; the CMS still needs a restart to reload a changed SKILL.md (it is cached per process).

  tutor dev restart cms

License

Proprietary.