- vlm_describer.py: objective extraction (beams, positions, GOSTs, dimensions) via qwen-vl-plus API. No error detection — only factual observation.
- vlm_qc_checker.py: VLM-based QC (deprecated in favor of rules-only QC)
- gost_dimension_validator.py: validate GOST references and dimension chains against known standards
- layout_detector.py: zone classification (drawing/table/title_block/notes) using line detection and text density analysis
- multi_element_extractor.py: extract dimensions, positions (П-1, X-1), GOST refs, steel grades, elevations, beam labels per zone
- tiling_ocr.py: split large drawings into overlapping tiles for better small-text recognition
- preprocess_for_ocr.py: CLAHE + unsharp mask for enhancing blueprint contrast
- visualize_dimensions.py: draw bounding boxes around detected dimension numbers
- compare_ocr.py: side-by-side visualization of normal vs tiling OCR results
- dimension_extractor.py: line-based dimension detection with pixel verification
- ocr_qwen.py: Alibaba Cloud qwen-vl-ocr client with resize and regex fallback parser
- test_qwen_ocr.py: standalone test for qwen OCR
- process_any_pdf.py: add --use-tiling flag to switch between normal and tiling OCR
