多语言 AI 平台怎么守住原文：七语保真、可控翻译与隐私架构实战

source_id -> source_hash -> source_language
          -> target_language
          -> translation
          -> engine_version
          -> glossary_version
          -> review_status

语言识别
  -> 输入安全检查
  -> 敏感级别与数据区域判断
  -> 术语表和上下文装配
  -> 翻译引擎路由
  -> 输出安全检查
  -> 原文一致性与格式校验
  -> 自动发布或人工复核

fastapi==0.115.0
uvicorn==0.30.6
pydantic==2.9.2

from datetime import datetime, timezone
from hashlib import sha256
from typing import Literal
from uuid import uuid4

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel, Field

app = FastAPI(title='Controlled Translation Gateway')

SUPPORTED = {
    'zh-CN', 'en-US', 'ja-JP', 'ko-KR',
    'fr-FR', 'de-DE', 'es-ES'
}

SOURCES: dict[str, dict] = {}
TRANSLATIONS: dict[str, dict] = {}


class TranslationRequest(BaseModel):
    text: str = Field(min_length=1, max_length=10000)
    source_language: str
    target_language: str
    sensitivity: Literal['public', 'internal', 'restricted'] = 'internal'
    glossary_version: str = 'default-v1'


def translate_adapter(text: str, target_language: str) -> str:
    # 可运行的占位实现；生产环境请替换为合规的翻译引擎。
    return f'[{target_language} machine draft] {text}'


def choose_route(sensitivity: str) -> str:
    # 示例策略：受限数据只能进入本地区域内的处理链路。
    if sensitivity == 'restricted':
        return 'in-region-private-engine'
    return 'approved-shared-engine'


@app.post('/v1/translations')
def create_translation(req: TranslationRequest):
    if req.source_language not in SUPPORTED:
        raise HTTPException(400, 'unsupported source language')
    if req.target_language not in SUPPORTED:
        raise HTTPException(400, 'unsupported target language')
    if req.source_language == req.target_language:
        raise HTTPException(400, 'source and target languages must differ')

    digest = sha256(req.text.encode('utf-8')).hexdigest()
    source_id = str(uuid4())
    translation_id = str(uuid4())
    now = datetime.now(timezone.utc).isoformat()

    # 原文单独存储；后续翻译不能写回这个对象。
    SOURCES[source_id] = {
        'source_id': source_id,
        'text': req.text,
        'language': req.source_language,
        'sha256': digest,
        'created_at': now,
    }

    route = choose_route(req.sensitivity)
    translated_text = translate_adapter(req.text, req.target_language)

    TRANSLATIONS[translation_id] = {
        'translation_id': translation_id,
        'source_id': source_id,
        'source_sha256': digest,
        'target_language': req.target_language,
        'translated_text': translated_text,
        'engine_route': route,
        'engine_version': 'adapter-demo-v1',
        'glossary_version': req.glossary_version,
        'review_status': 'machine_draft',
        'created_at': now,
    }

    return {
        'source': SOURCES[source_id],
        'translation': TRANSLATIONS[translation_id],
    }

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app:app --reload --port 8000

curl -s http://127.0.0.1:8000/v1/translations \
  -H 'Content-Type: application/json' \
  -d '{
    "text": "不得将客户数据传输到未批准的区域。",
    "source_language": "zh-CN",
    "target_language": "en-US",
    "sensitivity": "restricted",
    "glossary_version": "compliance-v3"
  }'

用户授权录音
  -> 端侧降噪或敏感片段裁剪
  -> 区域内临时加密存储
  -> 语音转写
  -> 文本脱敏
  -> 翻译与审核
  -> 按 TTL 删除原始音频