Imported from YUjinEDU/Lecture_Auto (lecture_auto/AGENTS.md). Install upstream with npx skills add YUjinEDU/Lecture_Auto --skill lecture_auto. Copyright stays with the author.

lecture_auto — backend package

Purpose

The Python backend. Contains the stage modules that do the real work (pipeline/) and the three orchestration layers that drive them: synchronous stages exposed to the CLI (run.py), async Celery tasks behind a FastAPI app (api/ + tasks/), and a self-contained local demo (demo/ + demo_static/). Cross-cutting concerns live in schemas/ (Pydantic models) and storage/ (disk-path conventions, voice files).

Key Files

File	Description
`__init__.py`	Package marker (minimal).

Subdirectories

Directory	Purpose
`pipeline/`	The shared stage core: parse → render → VLM → script → TTS → video (see `pipeline/AGENTS.md`).
`tasks/`	Celery tasks wrapping the stages with checkpointing + Redis progress (see `tasks/AGENTS.md`).
`api/`	FastAPI app: routes for jobs, scripts, TTS, download, demo (see `api/AGENTS.md`).
`demo/`	Self-contained demo orchestration with CPU fallbacks (see `demo/AGENTS.md`).
`demo_static/`	Vanilla-JS single-page UI for the demo, served by FastAPI (see `demo_static/AGENTS.md`).
`schemas/`	Pydantic v2 models shared across layers (manifest, script, tts, job status) (see `schemas/AGENTS.md`).
`storage/`	Per-job filesystem path helpers + voice-reference storage (see `storage/AGENTS.md`).

For AI Agents

Working In This Directory

Don't put orchestration logic in pipeline/. Stage modules must stay pure and driver-agnostic (callable from CLI, Celery, and demo alike). Sequencing, retries, progress, and persistence belong in tasks/, demo/, or run.py.
A stage's atomic unit is one slide (e.g. generate_single_note, synthesize_slide, regenerate_slide_task) so work is resumable.
Schemas in schemas/ are the contract between layers — change them deliberately and update both producers (pipeline) and consumers (api/tasks/demo + portal types).

Testing Requirements

Every module here has a matching tests/test_<module>.py. Run pytest; heavy externals are mocked.

Common Patterns

Pydantic v2 for all I/O; model_copy(update=...) for immutable edits.
Lazy model loading per worker (_get_vlm/_get_tts singletons) to avoid re-loading GPU weights.

Dependencies

Internal

All layers depend on pipeline/ and schemas/; tasks/ and api/ add Redis/Celery.

External

See root AGENTS.md. This package targets Python 3.11+.

lecture_auto

About