SELECT t.seqno, p.pc_seqno, t.name, t.xml_content, p.pod_type
FROM t_template t
JOIN t_product p ON t.product_seqno = p.seqno
WHERE p.pod_type = 'namecard'
  AND t.xml_content IS NOT NULL
  AND LENGTH(t.xml_content) > 0
ORDER BY t.seqno;

/data/dump/
├── templates_namecard.csv
├── templates_tshirt.csv
├── templates_leaflet.csv
└── ...

from lxml import etree
from urllib.parse import unquote
 
def parse_template(xml_content: str) -> etree._Element:
    if not xml_content.strip().startswith("<?xml"):
        xml_content = '<?xml version="1.0" encoding="UTF-8"?>' + xml_content
    return etree.fromstring(xml_content.encode("utf-8"))
 
def decode_color(value: int) -> str:
    """decimal color → hex 변환 (24-bit RGB / 32-bit ARGB)"""
    if value > 0xFFFFFF:
        r = (value >> 16) & 0xFF
        g = (value >> 8) & 0xFF
        b = value & 0xFF
        return f"#{r:02X}{g:02X}{b:02X}"
    return f"#{value:06X}"

from collections import Counter
 
def extract_metadata_by_code(root: etree._Element) -> dict:
    colors = set()
    for elem in root.iter():
        color_val = elem.get("color")
        if color_val and color_val.isdigit():
            colors.add(decode_color(int(color_val)))
 
    fonts = {elem.get("fontFamily") for elem in root.iter() if elem.get("fontFamily")}
 
    element_types = ["backgroundimage", "colorbox", "picturebox", "textbox"]
    composition = Counter(
        elem.tag for elem in root.iter() if elem.tag in element_types
    )
 
    return {
        "colors": sorted(colors),
        "fonts": sorted(fonts),
        "composition": dict(composition),
        "canvas_size": f"{root.get('width', '0')}x{root.get('height', '0')}",
    }

PII_HEAVY_TYPES = {"namecard", "card", "invitation", "envelope"}
 
async def mask_pii(text: str, pod_type: str) -> str:
    # 1단계: 코드 — 전화번호, 이메일 (모든 POD_TYPE, $0)
    text = mask_pii_by_code(text)
 
    # 2단계: AI — 이름, 주소 (PII 빈도 높은 유형만)
    if pod_type in PII_HEAVY_TYPES:
        text = await mask_pii_by_ai(text)
 
    return text

템플릿명: 심플 비즈니스 명함
생산타입: namecard
카테고리: 비즈니스 > 기업
색상: #465F67, #FFFFFF, #000000
폰트: NanumGothic, Arial
배경: 단색
요소: textbox:5, colorbox:3, picturebox:1
스타일: 미니멀, 모던, 클린
분위기: 깔끔하고 전문적인 비즈니스 느낌
용도: 비즈니스, 기업, 공식

async def embed_batch(texts: list[str]) -> list[list[float]]:
    response = await client.embeddings.create(
        model="text-embedding-3-large",
        input=texts,
        dimensions=3072,
    )
    return [item.embedding for item in response.data]

async def insert_vectors(templates: list[dict], vectors: list[list[float]]):
    records = [
        (t["seqno"], t["pod_type"], np.array(v, dtype=np.float16).tobytes())
        for t, v in zip(templates, vectors)
    ]
    async with pool.acquire() as conn:
        await conn.executemany(
            'UPDATE "T_TEMPLATE" SET "EMBEDDING" = $3::halfvec '
            'WHERE "SEQNO" = $1 AND "POD_TYPE" = $2',
            records,
        )

async def process_all_embeddings(templates, batch_size=2048):
    for i in range(0, len(templates), batch_size):
        batch = templates[i:i + batch_size]
        vectors = await embed_batch([t["embedding_text"] for t in batch])
        await insert_vectors(batch, vectors)
        await save_checkpoint(batch[-1]["seqno"])