haham7/nbc-chatbot

1"""2Lightweight cross-reference regex helpers (no llama_index dependency).3Kept in sync with ``NBCReferenceExtractor`` in ``retrieval_pipeline.py``.4"""5 6from __future__ import annotations7 8import re9from typing import Dict, List10 11 12class ReferencePatternExtractor:13    """Extract cross-references from NBC text using regex patterns."""14 15    CLAUSE_PATTERN = re.compile(16        r"(?:Clause|sec\.?|s\.)?\s*(\d+(?:\.\d+)*[a-z]?)", re.IGNORECASE17    )18    PART_PATTERN = re.compile(r"(?:Part|pt\.?)\s*(\d+)", re.IGNORECASE)19    TABLE_PATTERN = re.compile(r"(?:Table|tbl\.?)\s*(\d+(?:\.\d+)*)", re.IGNORECASE)20    FIGURE_PATTERN = re.compile(r"(?:Figure|Fig\.?)\s*(\d+(?:\.\d+)*)", re.IGNORECASE)21    FORMULA_PATTERN = re.compile(r"(?:Equation|eq\.?)\s*(\d+(?:\.\d+)*)", re.IGNORECASE)22    IS_PATTERN = re.compile(r"IS\s*:\s*(\d+)", re.IGNORECASE)23    NBC_PATTERN = re.compile(r"NBC\s*:\s*(\d+)", re.IGNORECASE)24 25    @classmethod26    def extract_all(cls, text: str) -> Dict[str, List[str]]:27        references = {28            "clauses": cls._normalize_clauses(cls.CLAUSE_PATTERN.findall(text)),29            "parts": cls.PART_PATTERN.findall(text),30            "tables": cls.TABLE_PATTERN.findall(text),31            "figures": cls.FIGURE_PATTERN.findall(text),32            "formulas": cls.FORMULA_PATTERN.findall(text),33            "is_codes": cls.IS_PATTERN.findall(text),34        }35        return references36 37    @classmethod38    def _normalize_clauses(cls, clauses: List[str]) -> List[str]:39        normalized = []40        for c in clauses:41            c = c.strip()42            if c and c[-1].isalpha() and not c[-2:].replace(".", "").isdigit():43                pass44            else:45                normalized.append(c.rstrip("abcdefghijklmnopqrstuvwxyz"))46        return [c for c in normalized if c] if normalized else clauses47 48    @classmethod49    def extract_clause_numbers(cls, text: str) -> List[str]:50        return cls.CLAUSE_PATTERN.findall(text)51