Function _build_pdf_chunks

agents/reading-agent/main.py:3214–3257

(parsed_pdf: Optional[Dict[str, Any]])

def _build_pdf_chunks(parsed_pdf: Optional[Dict[str, Any]]) -> List[Dict[str, Any]]:
    if not isinstance(parsed_pdf, dict):
        return []

    chunks: List[Dict[str, Any]] = []
    abstract = _clean_text(parsed_pdf.get("abstract"))
    if abstract:
        chunks.extend(_chunk_text_with_overlap(abstract, "abstract"))

    sections = parsed_pdf.get("sections") or {}
    if isinstance(sections, dict):
        preferred_order = [
            "introduction",
            "background",
            "method",
            "approach",
            "model",
            "results",
            "experiments",
            "evaluation",
            "discussion",
            "limitations",
            "conclusion",
        ]
        ordered_keys = preferred_order + [key for key in sections.keys() if key not in preferred_order]
        for section_name in ordered_keys:
            section_text = _clean_text(sections.get(section_name))
            if section_text:
                chunks.extend(_chunk_text_with_overlap(section_text, section_name))

    if not chunks:
        full_text = _clean_text(parsed_pdf.get("full_text"))
        if full_text:
            chunks.extend(_chunk_text_with_overlap(full_text, "full_text"))

    deduped: List[Dict[str, Any]] = []
    seen_texts = set()
    for chunk in chunks:
        key = chunk["text"].casefold()
        if key in seen_texts:
            continue
        seen_texts.add(key)
        deduped.append(chunk)
    return deduped


def _collect_evidence_sentences(
    evidence: Dict[str, Any],