init,llm gateway & import_analyse

app/services/import_analysis.py

@@ -0,0 +1,91 @@
+from __future__ import annotations
+
+from typing import List, Tuple
+
+from app.models import (
+    DataImportAnalysisRequest,
+    LLMMessage,
+    LLMProvider,
+    LLMRole,
+)
+
+
+def resolve_provider_from_model(llm_model: str) -> Tuple[LLMProvider, str]:
+    """Resolve provider based on the llm_model string.
+
+    The llm_model may be provided as 'provider:model' or 'provider/model'.
+    If no provider prefix is present, try an educated guess from common model name patterns.
+    """
+
+    normalized = llm_model.strip()
+    provider_hint: str | None = None
+    model_name = normalized
+
+    for delimiter in (":", "/", "|"):
+        if delimiter in normalized:
+            provider_hint, model_name = normalized.split(delimiter, 1)
+            provider_hint = provider_hint.strip().lower()
+            model_name = model_name.strip()
+            break
+
+    provider_map = {provider.value: provider for provider in LLMProvider}
+
+    if provider_hint:
+        if provider_hint not in provider_map:
+            raise ValueError(
+                f"Unsupported provider '{provider_hint}'. Expected one of: {', '.join(provider_map.keys())}."
+            )
+        return provider_map[provider_hint], model_name
+
+    return _guess_provider_from_model(model_name), model_name
+
+
+def _guess_provider_from_model(model_name: str) -> LLMProvider:
+    lowered = model_name.lower()
+
+    if lowered.startswith(("gpt", "o1", "text-", "dall-e", "whisper")):
+        return LLMProvider.OPENAI
+    if lowered.startswith(("claude", "anthropic")):
+        return LLMProvider.ANTHROPIC
+    if lowered.startswith(("gemini", "models/gemini")):
+        return LLMProvider.GEMINI
+    if lowered.startswith("qwen"):
+        return LLMProvider.QWEN
+    if lowered.startswith("deepseek"):
+        return LLMProvider.DEEPSEEK
+    if lowered.startswith(("openrouter", "router-")):
+        return LLMProvider.OPENROUTER
+
+    supported = ", ".join(provider.value for provider in LLMProvider)
+    raise ValueError(
+        f"Unable to infer provider from model '{model_name}'. "
+        f"Please prefix with 'provider:model'. Supported providers: {supported}."
+    )
+
+
+def build_import_messages(
+    request: DataImportAnalysisRequest,
+) -> List[LLMMessage]:
+    """Create system and user messages for the import analysis prompt."""
+    headers_formatted = "\n".join(f"- {header}" for header in request.table_headers)
+
+    system_prompt = (
+        "你是一名数据导入识别助手。请根据给定的表头和示例数据，判断字段含义、"
+        "典型数据类型以及潜在的数据质量问题。最终请返回一个结构化的JSON。\n"
+        "JSON结构需包含: field_summaries (数组, 每项含 header、meaning、data_type、quality_notes), "
+        "detected_issues (字符串数组)，以及 overall_suggestion (字符串)。"
+    )
+
+    user_prompt = (
+        f"导入记录ID: {request.import_record_id}\n\n"
+        "表头信息:\n"
+        f"{headers_formatted}\n\n"
+        "示例数据:\n"
+        f"{request.example_data}\n\n"
+        "请仔细分析示例数据与表头之间的对应关系，并返回符合上述JSON结构的内容。"
+    )
+
+    return [
+        LLMMessage(role=LLMRole.SYSTEM, content=system_prompt),
+        LLMMessage(role=LLMRole.USER, content=user_prompt),
+    ]