VTdevelops/bond-text-extraction

1from __future__ import annotations2 3from pathlib import Path4from typing import Iterable, Sequence5 6from .llm_client import BondInfoExtractor7from .models import BondRecord8from .pdf_loader import extract_text_from_pdf9from .xml_builder import build_bond_xml10 11 12class ExtractionPipeline:13    """High-level orchestration for PDF ingestion to XML output."""14 15    def __init__(16        self,17        model: str = "gpt-4.1-mini",18        extractor: BondInfoExtractor | None = None,19    ) -> None:20        self.extractor = extractor or BondInfoExtractor(model=model)21 22    def run(23        self,24        pdf_paths: Sequence[Path | str],25        *,26        extra_instructions: str | None = None,27        output_path: Path | str | None = None,28    ) -> tuple[list[BondRecord], str]:29        """Extract records into XML; optionally write XML to *output_path*."""30 31        normalised_paths = [Path(path) for path in pdf_paths]32        records: list[BondRecord] = []33 34        for path in normalised_paths:35            document_text = extract_text_from_pdf(path)36            # Query the LLM per document so each PDF generates its own bond payload.37            per_doc_records = self.extractor.extract_bonds(38                [document_text],39                extra_instructions,40            )41            records.extend(per_doc_records)42        xml_doc = build_bond_xml(records)43 44        if output_path is not None:45            Path(output_path).write_text(xml_doc, encoding="utf-8")46 47        return records, xml_doc48