v25.9.15.0 - Versioned Question Bank First

AI Server now has a bank-first model for managing question banks before an Open edX course exists:

Department → Subject → Chapter → Bank Version → Bank Release → Open edX Course Mapping

Production rule: 1 Bank Release = 1 Open edX Library. When learning material changes, create a new Bank Version and Release. Old courses keep their old release until an admin explicitly maps them to a newer release.

Main UI: /bank

Main API prefix: /api/question-bank-v2

See docs/V25_9_15_0_VERSIONED_QUESTION_BANK_FIRST.md.

AI Learning Server for Open edX CMS — v25.9.14.6 Native Ulmo ItemBank

Bản này giữ Stable Family + Hard Duplicate Guard, đồng thời tạo đúng native Problem Bank Beta (itembank) trên Open edX Ulmo.3. Mỗi Library V2 problem được đồng bộ thành course-local ProblemBlock bằng luồng Studio native và được xác minh trước khi báo hoàn tất.

Starter implementation for an Open edX AI Learning Check system.

This release includes the full roadmap implementation from v9 to v15:

• Modular backend foundation
• Alembic migration setup
• Real/mock Open edX connector adapter
• Production-ready RBAC path
• GPT-5 mini / local OpenAI-compatible model gateway
• Question Bank + Review workflow
• Open edX OLX export and publish endpoint
• Cost governance: estimate, quota, hard stop, usage log
• Multi-page Next.js UI
• Tests, CI, monitoring hooks and production compose

Run local demo

cp .env.example .env
docker compose up --build

Open:

• Frontend: http://localhost:3000
• API docs: http://localhost:8000/docs
• MinIO console: http://localhost:9001 (dev/demo only)

Important modes

Demo mode

USE_MOCK_OPENEDX=true
MOCK_LLM=true
AUTH_MODE=demo
AUTO_CREATE_TABLES=true

Real API-first mode

USE_MOCK_OPENEDX=false
MOCK_LLM=false
AUTH_MODE=jwt
AUTO_CREATE_TABLES=false
OPENAI_API_KEY=...
OPENEDX_BASE_URL=https://your-openedx.example.edu
OPENEDX_CLIENT_ID=...
OPENEDX_CLIENT_SECRET=...

Then run:

cd backend
alembic -c alembic.ini upgrade head

Pages

• /dashboard — statistics, governance, cost summary
• /sync — sync course content from Open edX
• /generate — manual/chunk generation mode
• /review — Teacher Review Queue
• /question-bank — all questions + filter/search/sort/edit
• /export — OLX preview/download and publish-to-Open-edX action
• /jobs — job monitor
• /settings — demo role/course settings

Docs

• docs/V9_TO_V15_RELEASE_NOTES.md
• docs/PRODUCTION_DEPLOYMENT.md
• docs/OPENEDX_REAL_INTEGRATION.md
• docs/AUTH_RBAC_PRODUCTION.md
• docs/API.md
• docs/FRONTEND_STRUCTURE.md
• docs/OPENEDX_EXPORT.md

Commit suggestion

git add .
git commit -m "feat: upgrade AI Open edX server from v9 to v15 production foundation"

v16-v19 status

This version adds production-oriented algorithms from the 10/10 plan:

• Course tree traversal from Open edX blocks
• HTML/transcript/PDF/PPTX extraction layer
• Hash-based changed content detection
• Chunking with source references
• Topic extraction and topic coverage allocation
• Batch generation with topic allocation
• Source grounding and duplicate detection hooks
• Friendlier multi-page UI for Sync and Generate

See docs/V16_TO_V19_RELEASE_NOTES.md.

v24 - Chapter/Module Libraries

Course tạo nhiều Library theo Chapter/Module. Khi AI sinh câu hỏi ở Unit/PDF/Video/HTML, backend tìm Chapter cha, import câu hỏi vào Library của Chapter đó và giữ source_node_id để Problem Bank trong Unit random/filter đúng nguồn. Xem docs/V24_CHAPTER_LIBRARIES.md.

v24.2 - Responses API Gateway

Bản này mặc định dùng OpenAI Responses API cho GPT thật.

Cấu hình nhanh trong .env hoặc /settings:

MODEL_PROVIDER=openai
OPENAI_MODEL=gpt-5-mini
OPENAI_API_MODE=responses
MOCK_LLM=false
OPENAI_API_KEY=sk-...

Trong /settings, admin có thể bấm Test GPT. Nếu OK, kết quả sẽ hiển thị openai_responses/gpt-5-mini qua responses.

Chat Completions cũ vẫn giữ dưới mode chat_legacy để fallback/local compatible.

v24.3 Accurate Cost Metering

• Estimate trước khi chạy dùng /v1/responses/input_tokens cho payload Responses thật.
• Safety factor chỉ áp dụng cho Estimate/Hard Stop.
• Actual cost dùng usage thật và không nhân safety factor.
• Có API GET /api/cost/pricing/realtime?model=...&refresh=true để fetch/cache giá OpenAI realtime-ish.