pdf

academic_doc_generator.core.pdf

PDF processing utilities (Docling + pypdf) with comprehensive type annotations.

extract_annotations_with_positions(pdf_path, ignore_source=True)

Extract annotations (comments/highlights) and their positions using pypdf.

Annotations are categorized for processing: - "llm": Regular comments sent to LLM for rewriting - "quelle": Source-related comments (counted only) - "language": Grammar/spelling comments (counted only) - "ignore": Special markers like "ab hier" (excluded from output)

Parameters:

Name	Type	Description	Default
pdf_path	str	Path to the PDF file.	required
ignore_source	bool	Whether to categorize source-related comments as "quelle". If False, they are treated as regular "llm" comments. Defaults to True.	True

Returns:

Type	Description
dict[int, list[AnnotationData]]	Tuple of (annotations, stats):
CommentStats	annotations: Dict mapping 0-based page indices to lists of annotations
tuple[dict[int, list[AnnotationData]], CommentStats]	stats: Comment category statistics

Example

annotations, stats = extract_annotations_with_positions("thesis.pdf") stats {'quelle': 5, 'language': 3, 'ignore': 1} annotations[0][]['category'] 'llm'

Source code in src/academic_doc_generator/core/pdf.py

def extract_annotations_with_positions(
    pdf_path: str, ignore_source: bool = True
) -> tuple[dict[int, list[AnnotationData]], CommentStats]:
    """Extract annotations (comments/highlights) and their positions using pypdf.

    Annotations are categorized for processing:
    - "llm": Regular comments sent to LLM for rewriting
    - "quelle": Source-related comments (counted only)
    - "language": Grammar/spelling comments (counted only)
    - "ignore": Special markers like "ab hier" (excluded from output)

    Args:
        pdf_path: Path to the PDF file.
        ignore_source: Whether to categorize source-related comments as "quelle".
            If False, they are treated as regular "llm" comments. Defaults to True.

    Returns:
        Tuple of (annotations, stats):
        - annotations: Dict mapping 0-based page indices to lists of annotations
        - stats: Comment category statistics

    Example:
        >>> annotations, stats = extract_annotations_with_positions("thesis.pdf")
        >>> stats
        {'quelle': 5, 'language': 3, 'ignore': 1}
        >>> annotations[0][0]['category']
        'llm'
    """
    if pdf_path.lower().endswith(".docx"):
        return {}, {"quelle": 0, "language": 0, "ignore": 0}

    reader = PdfReader(pdf_path)
    annotations: dict[int, list[AnnotationData]] = {}
    stats: CommentStats = {"quelle": 0, "language": 0, "ignore": 0}

    for idx, page in enumerate(reader.pages):
        page_annots: list[AnnotationData] = []
        if "/Annots" in page:
            for annot_ref in page["/Annots"]:
                annot = annot_ref.get_object()
                subtype = annot.get("/Subtype")
                rect = annot.get("/Rect")
                quadpoints = annot.get("/QuadPoints")
                content = annot.get("/Contents")

                if content:
                    # Filter by subtype to avoid picking up links, widgets, etc. as comments
                    # Common comment subtypes: /Text (sticky notes), /Highlight, /Underline, etc.
                    allowed_subtypes = [
                        "/Text",
                        "/Highlight",
                        "/Underline",
                        "/Squiggly",
                        "/StrikeOut",
                        "/FreeText",
                        "/Stamp",
                        "/Ink",
                        "/Line",
                        "/Square",
                        "/Circle",
                        "/Polygon",
                        "/PolyLine",
                    ]
                    if subtype not in allowed_subtypes:
                        continue

                    text = content.strip()
                    category: CommentCategory = "llm"  # default

                    # --- Categorize ---
                    # Check for "ab hier" first or footnote references (complete ignore)
                    if text.lower() == "ab hier" or is_footnote_reference(text):
                        category = "ignore"
                        stats["ignore"] += 1

                    # Check for "Quelle" comments
                    elif ignore_source and is_quelle_comment(text):
                        category = "quelle"
                        stats["quelle"] += 1

                    # Check for language-related comments
                    elif any(
                        kw in text.lower()
                        for kw in [
                            "rechtschreibung",
                            "grammatik",
                            "tippfehler",
                            "ausdruck",
                        ]
                    ):
                        category = "language"
                        stats["language"] += 1

                    page_annots.append(
                        {
                            "comment": text,
                            "subtype": subtype,
                            "rect": rect,
                            "quadpoints": quadpoints,
                            "category": category,
                        }
                    )

        if page_annots:
            annotations[idx] = page_annots

    return annotations, stats

extract_docx_text(docx_path)

Extract full text from a DOCX file including paragraphs and tables.

Parameters:

Name	Type	Description	Default
docx_path	str	Path to the DOCX file.	required

Returns:

Type	Description
str	The extracted plain text.

Source code in src/academic_doc_generator/core/pdf.py

def extract_docx_text(docx_path: str) -> str:
    """Extract full text from a DOCX file including paragraphs and tables.

    Args:
        docx_path: Path to the DOCX file.

    Returns:
        The extracted plain text.
    """
    import docx

    doc = docx.Document(docx_path)
    text_parts = []
    for para in doc.paragraphs:
        if para.text.strip():
            text_parts.append(para.text)
    for table in doc.tables:
        for row in table.rows:
            row_text = []
            for cell in row.cells:
                if cell.text.strip():
                    row_text.append(cell.text.strip())
            if row_text:
                text_parts.append(" | ".join(row_text))
    return "\n".join(text_parts)

extract_text_per_page(pdf_path, max_pages=10)

Extract plain text (without positions) for the first max_pages pages using LiteParse with docling and PyMuPDF fallbacks.

This is faster than extracting word positions and is sufficient for metadata extraction and thesis summarization. Supports Word documents (.docx) as well.

Parameters:

Name	Type	Description	Default
pdf_path	str	Path to the PDF file.	required
max_pages	Optional[int]	Maximum number of pages to read. If None, all pages are read. Defaults to 10.	10

Returns:

Type	Description
dict[int, str]	Dictionary mapping 0-based page indices to the full concatenated text
dict[int, str]	of that page.

Example

text = extract_text_per_page("thesis.pdf", max_pages=2) text[0][:50] 'Introduction This thesis examines the impact of...'

Source code in src/academic_doc_generator/core/pdf.py

def extract_text_per_page(pdf_path: str, max_pages: Optional[int] = 10) -> dict[int, str]:
    """Extract plain text (without positions) for the first `max_pages` pages using LiteParse with docling and PyMuPDF fallbacks.

    This is faster than extracting word positions and is sufficient for metadata
    extraction and thesis summarization. Supports Word documents (.docx) as well.

    Args:
        pdf_path: Path to the PDF file.
        max_pages: Maximum number of pages to read. If None, all pages are read.
            Defaults to 10.

    Returns:
        Dictionary mapping 0-based page indices to the full concatenated text
        of that page.

    Example:
        >>> text = extract_text_per_page("thesis.pdf", max_pages=2)
        >>> text[0][:50]
        'Introduction This thesis examines the impact of...'
    """
    import os

    if pdf_path.lower().endswith(".docx"):
        try:
            full_text = extract_docx_text(pdf_path)
            return {0: full_text}
        except Exception as e:
            print(f"❌ Error reading DOCX file: {e}")
            return {}

    is_valid_pdf = True
    if os.path.exists(pdf_path):
        with open(pdf_path, "rb") as f:
            header = f.read(4)
            if header != b"%PDF":
                is_valid_pdf = False

    if is_valid_pdf:
        try:
            import liteparse

            parser = liteparse.LiteParse()
            pdf_doc = parser.parse(pdf_path)
            pages_text: dict[int, str] = {}
            for idx, page in enumerate(pdf_doc.pages):
                if max_pages is not None and idx >= max_pages:
                    break
                pages_text[idx] = page.text or ""

            if any(text.strip() for text in pages_text.values()):
                return pages_text
            else:
                print("⚠️  LiteParse returned empty text. Falling back to Docling...")
        except (Exception, MemoryError) as e:
            print(f"⚠️  LiteParse failed ({type(e).__name__}: {e}). Falling back to Docling...")

    if is_valid_pdf:
        try:
            docling_parser = DoclingPdfParser()
            doc_pdf_doc = docling_parser.load(path_or_stream=pdf_path)

            pages_text = {}
            for zero_idx, (_page_no, pred_page) in enumerate(doc_pdf_doc.iterate_pages(), start=0):
                if max_pages is not None and zero_idx >= max_pages:
                    break
                words = [cell.text for cell in pred_page.iterate_cells(unit_type=TextCellUnit.WORD)]
                page_text = " ".join(words)
                pages_text[zero_idx] = page_text

            if any(text.strip() for text in pages_text.values()):
                return pages_text
            else:
                print("⚠️  Docling returned empty text. Falling back to PyMuPDF...")
        except (Exception, MemoryError) as doc_err:
            print(
                f"⚠️  Docling parser failed ({type(doc_err).__name__}: {doc_err}). Falling back to PyMuPDF..."
            )

    try:
        import pymupdf

        doc = pymupdf.open(pdf_path)
        pages_text = {}
        for page_idx, page in enumerate(doc):
            if max_pages is not None and page_idx >= max_pages:
                break
            words = [w[4] for w in page.get_text("words")]
            pages_text[page_idx] = " ".join(words)
        return pages_text
    except Exception as fallback_err:
        print(f"❌ LiteParse, Docling, and PyMuPDF fallback failed to parse PDF: {fallback_err}")
        return {}

extract_text_with_positions(pdf_path)

Extract text and bounding boxes for words from a PDF using LiteParse with docling and PyMuPDF fallbacks.

Parameters:

Name	Type	Description	Default
pdf_path	str	Path to the PDF file.	required

Returns:

Type	Description
dict[int, list[WordBox]]	Dictionary mapping 0-based page indices to a list of words with bounding boxes.

Example

words = extract_text_with_positions("thesis.pdf") words[0][]

Source code in src/academic_doc_generator/core/pdf.py

def extract_text_with_positions(pdf_path: str) -> dict[int, list[WordBox]]:
    """Extract text and bounding boxes for words from a PDF using LiteParse with docling and PyMuPDF fallbacks.

    Args:
        pdf_path: Path to the PDF file.

    Returns:
        Dictionary mapping 0-based page indices to a list of words with bounding boxes.

    Example:
        >>> words = extract_text_with_positions("thesis.pdf")
        >>> words[0][0]
        {'text': 'Introduction', 'bbox': (72.0, 720.0, 150.0, 735.0)}
    """
    import os

    if pdf_path.lower().endswith(".docx"):
        return {0: []}

    is_valid_pdf = True
    if os.path.exists(pdf_path):
        with open(pdf_path, "rb") as f:
            header = f.read(4)
            if header != b"%PDF":
                is_valid_pdf = False

    if is_valid_pdf:
        try:
            import liteparse

            parser = liteparse.LiteParse(emit_word_boxes=True)
            pdf_doc = parser.parse(pdf_path)
            pages_words: dict[int, list[WordBox]] = {}
            for idx, page in enumerate(pdf_doc.pages):
                words: list[WordBox] = []
                page_height = page.height
                for item in page.text_items:
                    if item.words:
                        for w in item.words:
                            x0 = float(w.x)
                            y0 = float(page_height - (w.y + w.height))
                            x1 = float(w.x + w.width)
                            y1 = float(page_height - w.y)
                            words.append({"text": w.text, "bbox": (x0, y0, x1, y1)})
                    else:
                        x0 = float(item.x)
                        y0 = float(page_height - (item.y + item.height))
                        x1 = float(item.x + item.width)
                        y1 = float(page_height - item.y)
                        parts = item.text.split()
                        if len(parts) <= 1:
                            words.append({"text": item.text, "bbox": (x0, y0, x1, y1)})
                        else:
                            word_width = item.width / len(parts)
                            for i, part in enumerate(parts):
                                px0 = x0 + i * word_width
                                px1 = px0 + word_width
                                words.append({"text": part, "bbox": (px0, y0, px1, y1)})
                pages_words[idx] = words

            if any(words for words in pages_words.values()):
                return pages_words
            else:
                print("⚠️  LiteParse returned empty text. Falling back to Docling...")
        except (Exception, MemoryError) as e:
            print(f"⚠️  LiteParse failed ({type(e).__name__}: {e}). Falling back to Docling...")

    if is_valid_pdf:
        try:
            docling_parser = DoclingPdfParser()
            doc_pdf_doc = docling_parser.load(path_or_stream=pdf_path)

            pages_words = {}

            # Enumerate to force 0-based indexing, regardless of docling's page_no (1-based)
            for zero_idx, (_page_no, pred_page) in enumerate(doc_pdf_doc.iterate_pages(), start=0):
                words = []
                for cell in pred_page.iterate_cells(unit_type=TextCellUnit.WORD):
                    r = (
                        cell.rect
                    )  # BoundingRectangle with r_x0, r_y0, r_x1, r_y1 (bottom-left origin)

                    words.append(
                        {
                            "text": cell.text,
                            "bbox": (
                                float(r.r_x0),
                                float(r.r_y0),
                                float(r.r_x1),
                                float(r.r_y1),
                            ),
                        }
                    )
                pages_words[zero_idx] = words

            if any(words for words in pages_words.values()):
                return pages_words
            else:
                print("⚠️  Docling returned empty text. Falling back to PyMuPDF...")
        except (Exception, MemoryError) as doc_err:
            print(
                f"⚠️  Docling parser failed ({type(doc_err).__name__}: {doc_err}). Falling back to PyMuPDF..."
            )

    try:
        import pymupdf

        doc = pymupdf.open(pdf_path)
        pages_words = {}
        for page_idx, page in enumerate(doc):
            rect = page.rect
            page_height = rect.y1
            words_list = page.get_text("words")
            words = []
            for w in words_list:
                # w is (x0, y0, x1, y1, text, block_no, line_no, word_no)
                x0, y0, x1, y1, text = w[0], w[1], w[2], w[3], w[4]
                words.append(
                    {
                        "text": text,
                        "bbox": (
                            float(x0),
                            float(page_height - y1),
                            float(x1),
                            float(page_height - y0),
                        ),
                    }
                )
            pages_words[page_idx] = words
        return pages_words
    except Exception as fallback_err:
        print(f"❌ LiteParse, Docling, and PyMuPDF fallback failed to parse PDF: {fallback_err}")
        return {}

find_annotation_context(pages_words, annotations)

Match annotations to the words and paragraphs they reference.

For each annotation, this function: 1. Finds the words that overlap with the annotation's bounding box 2. Extracts the highlighted text from those words 3. Finds the paragraph containing the highlighted text 4. Returns all context information for LLM processing

Parameters:

Name	Type	Description	Default
pages_words	dict[int, list[WordBox]]	Words with bounding boxes per page (0-based indices).	required
annotations	dict[int, list[AnnotationData]]	Annotations per page with rects and comments (0-based indices).	required

Returns:

Type	Description
dict[int, list[AnnotationContext]]	Dictionary mapping 1-based page numbers to lists of annotation contexts.
dict[int, list[AnnotationContext]]	Page numbers are 1-based for user display purposes.

Example

pages_words = {0: [{'text': 'test', 'bbox': (10, 10, 50, 20)}]} annotations = {0: [{'comment': 'Why?', 'rect': [5, 5, 55, 25], ... 'category': 'llm', 'subtype': '/Text', 'quadpoints': None}]} context = find_annotation_context(pages_words, annotations) context[1][0]['highlighted'] 'test'

Source code in src/academic_doc_generator/core/pdf.py

def find_annotation_context(
    pages_words: dict[int, list[WordBox]], annotations: dict[int, list[AnnotationData]]
) -> dict[int, list[AnnotationContext]]:
    """Match annotations to the words and paragraphs they reference.

    For each annotation, this function:
    1. Finds the words that overlap with the annotation's bounding box
    2. Extracts the highlighted text from those words
    3. Finds the paragraph containing the highlighted text
    4. Returns all context information for LLM processing

    Args:
        pages_words: Words with bounding boxes per page (0-based indices).
        annotations: Annotations per page with rects and comments (0-based indices).

    Returns:
        Dictionary mapping 1-based page numbers to lists of annotation contexts.
        Page numbers are 1-based for user display purposes.

    Example:
        >>> pages_words = {0: [{'text': 'test', 'bbox': (10, 10, 50, 20)}]}
        >>> annotations = {0: [{'comment': 'Why?', 'rect': [5, 5, 55, 25],
        ...                     'category': 'llm', 'subtype': '/Text', 'quadpoints': None}]}
        >>> context = find_annotation_context(pages_words, annotations)
        >>> context[1][0]['highlighted']
        'test'
    """
    context_dict: dict[int, list[AnnotationContext]] = {}

    for page_num, annots in annotations.items():
        page_results: list[AnnotationContext] = []

        for annot in annots:
            rect = annot["rect"]
            if rect:
                annot_bbox = tuple(rect)
            elif annot["quadpoints"]:
                qp = annot["quadpoints"]
                xs = qp[0::2]
                ys = qp[1::2]
                annot_bbox = (min(xs), min(ys), max(xs), max(ys))
            else:
                continue

            # Words under the annotation (with neighbor-page fallback)
            page_idx_for_words, hit_words = get_words_for_annotation_on_page(
                pages_words,
                page_num,
                annot_bbox,  # type: ignore[arg-type]
            )
            highlighted_text = " ".join([w["text"] for w in hit_words]).strip()

            # Use the full text of the page where words were actually found
            full_page_text = " ".join([w["text"] for w in pages_words.get(page_idx_for_words, [])])
            paragraphs = re.split(r"\n\s*\n| {2,}", full_page_text)

            # Find paragraph containing the highlighted words
            para_match: Optional[str] = None
            for para in paragraphs:
                if highlighted_text and highlighted_text in para:
                    para_match = para
                    break

            if not para_match and paragraphs:
                para_match = paragraphs[0]  # fallback

            page_results.append(
                {
                    "comment": annot["comment"],
                    "highlighted": highlighted_text,
                    "paragraph": para_match if para_match else "",
                    "category": annot.get("category", "llm"),
                }
            )

        if page_results:
            # +1 so reported page number is human-readable (1-based)
            context_dict[page_num + 1] = page_results

    return context_dict

get_words_for_annotation_on_page(pages_words, page_index, rect)

Get words that match an annotation rectangle, checking neighboring pages if necessary.

Sometimes annotations appear on the wrong page in the PDF structure. This function checks the specified page first, then the next page (+1), then the previous page (-1).

Parameters:

Name	Type	Description	Default
pages_words	dict[int, list[WordBox]]	Dictionary of pages mapped to word lists.	required
page_index	int	Index of the annotated page (0-based).	required
rect	tuple[float, float, float, float]	Annotation rectangle as (x0, y0, x1, y1).	required

Returns:

Type	Description
int	Tuple of (page_index_used, words) where page_index_used is the page
list[WordBox]	where words were actually found, and words is the list of matching word dicts.

Example

pages_words = {0: [{'text': 'test', 'bbox': (10, 10, 50, 20)}]} rect = (5, 5, 55, 25) page_idx, words = get_words_for_annotation_on_page(pages_words, 0, rect) page_idx 0 len(words) 1

Source code in src/academic_doc_generator/core/pdf.py

def get_words_for_annotation_on_page(
    pages_words: dict[int, list[WordBox]],
    page_index: int,
    rect: tuple[float, float, float, float],
) -> tuple[int, list[WordBox]]:
    """Get words that match an annotation rectangle, checking neighboring pages if necessary.

    Sometimes annotations appear on the wrong page in the PDF structure. This function
    checks the specified page first, then the next page (+1), then the previous page (-1).

    Args:
        pages_words: Dictionary of pages mapped to word lists.
        page_index: Index of the annotated page (0-based).
        rect: Annotation rectangle as (x0, y0, x1, y1).

    Returns:
        Tuple of (page_index_used, words) where page_index_used is the page
        where words were actually found, and words is the list of matching word dicts.

    Example:
        >>> pages_words = {0: [{'text': 'test', 'bbox': (10, 10, 50, 20)}]}
        >>> rect = (5, 5, 55, 25)
        >>> page_idx, words = get_words_for_annotation_on_page(pages_words, 0, rect)
        >>> page_idx
        0
        >>> len(words)
        1
    """
    # Try the given page, then +1, then -1
    candidates = [page_index, page_index + 1, page_index - 1]
    for idx in candidates:
        if idx in pages_words:
            hits = words_overlapping_rect(pages_words[idx], rect)
            if hits:
                return idx, hits
    # fall back to the original page even if empty
    return page_index, []

is_footnote_reference(text)

Check if the text is likely just a footnote reference or citation marker.

Parameters:

Name	Type	Description	Default
text	str	The text to check.	required

Returns:

Type	Description
bool	True if it looks like a footnote or citation reference, False otherwise.

Examples:

>>> is_footnote_reference("[1]")
True
>>> is_footnote_reference("1.")
True
>>> is_footnote_reference("[Mül23]")
True

Source code in src/academic_doc_generator/core/pdf.py

def is_footnote_reference(text: str) -> bool:
    """Check if the text is likely just a footnote reference or citation marker.

    Args:
        text: The text to check.

    Returns:
        True if it looks like a footnote or citation reference, False otherwise.

    Examples:
        >>> is_footnote_reference("[1]")
        True
        >>> is_footnote_reference("1.")
        True
        >>> is_footnote_reference("[Mül23]")
        True
    """
    normalized = text.strip()
    # matches [1], [12], 1., (1)
    # or typical German citation style [Mül23] or [ABC23]
    return bool(
        re.match(r"^(\[?\d+\]?|\d+\.|\(\d+\))$", normalized)
        or re.match(r"^\[[A-Z][a-zäöüA-Z]{1,3}\d{2}\]$", normalized)
    )

is_quelle_comment(text, max_length=20)

Check if a comment is a source-related comment that should be counted but not rewritten.

Source comments (containing "Quelle" or "source") are counted in statistics but not sent to the LLM for rewriting. They must be short (≤max_length characters) and contain the keyword as a whole word (not part of another word like "Consequent").

Parameters:

Name	Type	Description	Default
text	str	The comment text to check.	required
max_length	int	Maximum length for a comment to be considered a Quelle comment. Defaults to 20 characters.	20

Returns:

Type	Description
bool	True if this is a source-related comment, False otherwise.

Examples:

>>> is_quelle_comment("Quelle?")
True
>>> is_quelle_comment("Source missing")
True
>>> is_quelle_comment("Consequent")  # Not a whole word match
False
>>> is_quelle_comment("Quelle fehlt hier an dieser Stelle komplett")
False  # Too long

Source code in src/academic_doc_generator/core/pdf.py

def is_quelle_comment(text: str, max_length: int = 20) -> bool:
    """Check if a comment is a source-related comment that should be counted but not rewritten.

    Source comments (containing "Quelle" or "source") are counted in statistics but not
    sent to the LLM for rewriting. They must be short (≤max_length characters) and contain
    the keyword as a whole word (not part of another word like "Consequent").

    Args:
        text: The comment text to check.
        max_length: Maximum length for a comment to be considered a Quelle comment.
            Defaults to 20 characters.

    Returns:
        True if this is a source-related comment, False otherwise.

    Examples:
        >>> is_quelle_comment("Quelle?")
        True
        >>> is_quelle_comment("Source missing")
        True
        >>> is_quelle_comment("Consequent")  # Not a whole word match
        False
        >>> is_quelle_comment("Quelle fehlt hier an dieser Stelle komplett")
        False  # Too long
    """
    # Normalize text
    normalized = text.strip().lower()

    # Check length constraint
    if len(normalized) > max_length:
        return False

    # Check for "quelle" or "source" keywords using regex word boundaries
    quelle_pattern = r"\bquelle\b|\bsource\b"
    return bool(re.search(quelle_pattern, normalized, re.IGNORECASE))

rect_overlap(word_bbox, annot_bbox)

Check if a word bounding box overlaps with an annotation rectangle.

Parameters:

Name	Type	Description	Default
word_bbox	tuple[float, float, float, float]	Word bounding box as (x0, y0, x1, y1).	required
annot_bbox	tuple[float, float, float, float]	Annotation bounding box as (x0, y0, x1, y1).	required

Returns:

Type	Description
bool	True if the bounding boxes overlap, False otherwise.

Example

rect_overlap((10, 10, 50, 20), (5, 5, 55, 25)) True rect_overlap((10, 10, 50, 20), (100, 100, 150, 120)) False

Source code in src/academic_doc_generator/core/pdf.py

def rect_overlap(
    word_bbox: tuple[float, float, float, float],
    annot_bbox: tuple[float, float, float, float],
) -> bool:
    """Check if a word bounding box overlaps with an annotation rectangle.

    Args:
        word_bbox: Word bounding box as (x0, y0, x1, y1).
        annot_bbox: Annotation bounding box as (x0, y0, x1, y1).

    Returns:
        True if the bounding boxes overlap, False otherwise.

    Example:
        >>> rect_overlap((10, 10, 50, 20), (5, 5, 55, 25))
        True
        >>> rect_overlap((10, 10, 50, 20), (100, 100, 150, 120))
        False
    """
    x1, y1, x2, y2 = word_bbox
    ax1, ay1, ax2, ay2 = annot_bbox
    return not (x2 < ax1 or x1 > ax2 or y2 < ay1 or y1 > ay2)

words_overlapping_rect(words, rect, tol=0.5)

Find all words that overlap with a given rectangle.

Parameters:

Name	Type	Description	Default
words	list[WordBox]	List of word dictionaries with 'text' and 'bbox'.	required
rect	tuple[float, float, float, float]	Annotation rectangle as (x0, y0, x1, y1).	required
tol	float	Tolerance factor for overlap detection in points. Defaults to 0.5.	0.5

Returns:

Type	Description
list[WordBox]	List of words that overlap with the rectangle.

Example

words = [{'text': 'Hello', 'bbox': (10, 10, 50, 20)}] rect = (5, 5, 55, 25) overlapping = words_overlapping_rect(words, rect) len(overlapping) 1

Source code in src/academic_doc_generator/core/pdf.py

def words_overlapping_rect(
    words: list[WordBox], rect: tuple[float, float, float, float], tol: float = 0.5
) -> list[WordBox]:
    """Find all words that overlap with a given rectangle.

    Args:
        words: List of word dictionaries with 'text' and 'bbox'.
        rect: Annotation rectangle as (x0, y0, x1, y1).
        tol: Tolerance factor for overlap detection in points. Defaults to 0.5.

    Returns:
        List of words that overlap with the rectangle.

    Example:
        >>> words = [{'text': 'Hello', 'bbox': (10, 10, 50, 20)}]
        >>> rect = (5, 5, 55, 25)
        >>> overlapping = words_overlapping_rect(words, rect)
        >>> len(overlapping)
        1
    """
    x0, y0, x1, y1 = rect
    hits: list[WordBox] = []
    for w in words:
        wx0, wy0, wx1, wy1 = w["bbox"]
        if wx1 >= x0 - tol and wx0 <= x1 + tol and wy1 >= y0 - tol and wy0 <= y1 + tol:
            hits.append(w)
    return hits