4 Commits

Author	SHA1	Message	Date
Woody	ef10b937cf	feat: Sub-Phase 8.0 — config & enums for Q&A-pair chunking strategy ... Backend: - Add 6 Q&A chunking config fields to Settings (default_chunking_strategy, qa_vision_enabled, qa_max_chunk_tokens, qa_structure_model, qa_include_internal_refs, qa_cache_vision_results) - Define ChunkingStrategyType Literal + VALID_CHUNKING_STRATEGIES frozenset - Add strategy field to IngestResponse (default token, non-breaking) - Add IngestRequest model with strategy param - Update .env.example with new env vars Frontend: - Add ChunkingStrategy type ('token' | 'question') - Extend IngestResponse, DocumentInfo, ChunkInfo with Q&A fields Tests: - test_qa_chunking_config_defaults — all defaults verified - test_qa_chunking_config_from_env — env var overrides verified Plan fix: renamed qa_verification_model → qa_structure_model to match LLM-first architecture	2026-05-15 12:01:28 +08:00
Woody	6bf04cedb1	docs: Package 8 — switch to LLM-first structure detection (not regex-first) ... LegCo documents use multiple formats (問/答 markers, Q1/Q2 numbering, section headings like '(1) 住戶的安置補償', 發言要點 bullet points, and pure table pages). Regex alone cannot reliably classify all these. Changes: - Primary detection: LLM call identifies ALL section types in one pass (qa, narrative, speaking_notes, table, toc, heading_only) - Regex: downgraded to optional fast-pass optimization for known patterns - Architecture diagram, algorithm detail, risks, and test plan all updated - Single model handles structure detection + table extraction + verification	2026-05-15 11:34:24 +08:00
Woody	322caf1cc0	docs: Package 8 — add vLLM vision compatibility risk and smoke test to plan ... - New risk: vLLM may not support Qwen3.5-35B-A3B vision API depending on version - Dependencies: added vLLM compatibility note with smoke test snippet - Heuristic fallback (Option B) works regardless of OpenRouter or vLLM - qa_vision_enabled toggle provides escape hatch	2026-05-15 11:20:20 +08:00
Woody	dbae9411c6	docs: Package 8 enhancement plan — Q&A-pair chunking strategy with vision table extraction ... - New QuestionChunkingStrategy splits by 問/答 and Q1/Q2 boundaries - Vision-based table-to-markdown using existing Qwen3.5-35B-A3B (native vision model) - Strategy selector UI on RAG Database page (token vs question) - Hybrid approach: regex primary split + LLM verification for edge cases - Single-model architecture — no separate vision API needed - 6 sub-phases with test-first delivery, 7 new files, 15+ modified files	2026-05-15 11:10:36 +08:00