意图识别与 RAG

1. 登录 AIP，打开"智能查询"
2. 输入"每个客户的订单数量"
3. 发送并等待返回
4. 查看 SQL 说明、图表建议与数据表，以及 rag_channels_used 字段

{
  "intent": "data_query", "confidence": 0.8,
  "sql_query": "SELECT c.customer_name, COUNT(o.order_id) AS cnt FROM orders o JOIN customers c ON o.customer_id = c.customer_id GROUP BY c.customer_name",
  "visualization": { "chart_type": "bar", "reason": "one dimension and numeric columns" },
  "query_result": { "columns": ["customer_name", "cnt"], "rows": [["张三", 2], ["李四", 2], ["王五", 1]] },
  "rag_channels_used": ["metadata", "knowledge"]
}

1. 打开"智能查询"
2. 输入"什么是订单"（或"解释一下销售额"）
3. 发送，观察返回（不进入 RAG / Text2SQL / 执行链路）

{
  "intent": "knowledge_qa", "confidence": 0.9,
  "message": "知识问答类问题暂由知识库处理，当前最小链路仅支持数据查询与分析。",
  "data_source_name": "aip_demo_warehouse"
}

1. 打开"智能查询"，输入"帮我看看最近谁买得多"
2. 发送，观察 intent / confidence / rag_context / rag_channels_used
3. 查看 SQL 说明，判断模型理解成了"订单数"还是"销售金额"

{
  "intent": "data_query", "confidence": 0.5,
  "rag_channels_used": [],
  "rag_context": "Table: orders\n  - order_id (INTEGER)\n  - sales_amount (REAL)\n  - customer_id (INTEGER)\n...",
  "sql_query": "SELECT c.customer_name, SUM(o.sales_amount) AS total FROM orders o JOIN customers c ON o.customer_id = c.customer_id GROUP BY c.customer_name ORDER BY total DESC"
}

1. 登录 admin，确认数据源 aip_demo_warehouse 已注册
2. 调用 POST /api/v1/datasources/:id/import-metadata?ai_generate_descriptions=true
3. 调用 POST /api/v1/rag/index/rebuild 重建元数据向量索引
4. 用几条真实问法回归测试，看 rag_channels_used / rag_context 命中情况
5. 对比优化前后的命中率

-- import-metadata 返回（示意）
{ "status": "success", "message": "Metadata import complete. Created 0 tables and 0 columns. Updated 3 tables and 12 columns." }
-- index/rebuild 返回（示意）
{ "code": 0, "data": { "data_source": "aip_demo_warehouse", "items": 15, "indexed": 15, "failed": 0, "degraded": false, "collection": "metadata_embeddings" } }
-- 优化后 RAG 上下文（不再回退全量）
Table: orders
  - sales_amount (REAL) (销售额)
Table: customers
  - customer_name (TEXT) (客户名)