Esta web es mi demo

def window(text: str, max_chars: int, overlap: int) -> list[str]:
    """Corta textos largos en ventanas, intentando terminar en un párrafo o frase."""
    if len(text) <= max_chars:
        return [text]
    if overlap >= max_chars:
        raise ValueError("overlap debe ser menor que max_chars")
    parts, start = [], 0
    while start < len(text):
        end = min(start + max_chars, len(text))
        if end < len(text):
            cut = max(text.rfind("\n\n", start, end), text.rfind(". ", start, end))
            if cut > start + max_chars // 2:
                end = cut + 1
        parts.append(text[start:end].strip())
        if end >= len(text):
            break
        start = max(end - overlap, start + 1)
    return [p for p in parts if p]

def load_previous_vectors(index_path: Path, model: str) -> dict[str, list[float]]:
    """Carga incremental: reutiliza embeddings de fragmentos cuyo texto no cambió."""
    if not index_path.exists():
        return {}
    try:
        previous = json.loads(index_path.read_text(encoding="utf-8"))
    except json.JSONDecodeError:
        log.warning("Índice anterior ilegible: se regenerará completo")
        return {}
    if previous.get("model") != model:
        log.info("Cambió el modelo de embeddings: se regenerará completo")
        return {}
    return {c["hash"]: c["embedding"] for c in previous.get("chunks", []) if c.get("embedding")}