Un singur set de scripturi acum rulează pe orice curs configurat în
courses.py. Master rămâne la rădăcina repo (backward-compat M1-M6);
cursuri noi (ex. practitioner la shop.cursnlp.ro) primesc un root
dedicat (nlp-practitioner/) cu propriile artefacte.
- courses.py: config dict (master, practitioner) + course_paths() +
validate_manifest_course() (manifest fără course_key = master).
- download.py: --course + --modules; trei tipuri de lecții (audio HTTP,
Vimeo iframe via yt-dlp audio-only, text-only cu captură HTML);
merge cu manifest existent în loc de replace; strip [Audio] pentru
backward-compat paths.
- transcribe.py: --course + --modules; skip type==text; path-uri prin
course_paths(); validare course_key.
- summarize.py: --course + --compile; template prompt folosește
course['name']; scrie SUPORT_CURS.md cu LF explicit (WSL2 baseline).
- md_to_pdf.py: --course resolv-ă summaries_dir / pdf_dir per curs.
- run.bat: detectează master|practitioner ca primul argument,
propagă --course la sub-scripturi; backward-compat run.bat [modules].
- requirements.txt: + yt-dlp.
- .gitignore: nlp-practitioner/audio/, audio_wav/, scratch_recon.py, tmp_recon/.
- tests/test_regression.sh: 5 gate-uri read-only (import, schema,
disk-coherence, SUPORT_CURS byte-identic, cross-course isolation).
Regression curs master: PASS (manifest + SUPORT_CURS.md hash
identic cu baseline /tmp/suport_before.md).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- transcribe.py: add --max-context 0, --entropy-thold 2.4, --max-len 60,
--suppress-nst, --no-fallback to whisper.cpp to prevent hallucination loops
- transcribe.py: remove interactive quality gate (runs unattended now)
- run.bat: remove pause prompts for unattended operation
- retranscribe_tail.py: new script that detects hallucination bursts in SRT
files, extracts and re-transcribes only the affected audio segments, then
splices the result back together. Drops segments that re-hallucinate
(silence/music). Backs up originals to transcripts/backup/.
- fix_hallucinations.bat: Windows wrapper for retranscribe_tail.py
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- run.bat: one-click pipeline (download, convert, transcribe)
- download.py: fetch audio from course platform
- transcribe.py: whisper.cpp batch transcription (CPU, WAV 16kHz)
- MP3->WAV conversion via ffmpeg
- --modules filter for splitting work across machines
- summarize.py: generate summaries from transcripts
- setup_whisper.py: auto-download whisper.cpp, ffmpeg, and model
- Medium model (q5_0) instead of large to avoid VRAM crashes
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
