joungmin d0c2aa3857 feat: replace single-pass enricher with 4-step pipeline ...

Upgrades content processing from a single LLM call to a structured
5-step document reconstruction pipeline:

  1. Normalize  — 구어체 정제, 문장부호 복원, 핵심 엔티티 추출
  2. Index Tree — 텍스트 전체 스캔 → 계층적 목차(JSON) 생성
  3. Leaf Summarize — 섹션별 상세 요약 (context overlap 300자 적용)
  4. Consistency Check — 누락 엔티티 검증 및 보완
  5. Assemble — 최종 Markdown 문서 조립 (LLM 불필요)

- Short texts (< 3000 chars): simple 1-pass fallback
- Long texts: full pipeline (N+4 LLM calls where N = section count)
- worker.py: uses body_md from enricher as Obsidian note body

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-03-02 18:02:00 +09:00

3.6 KiB

Raw Permalink Blame History

View Raw