Fix live multi-turn: pass the resolved drug, stop did-you-mean garbage

apps/ai-service/config.py

@@ -26,8 +26,14 @@ class Settings(BaseSettings):
     aws_region: str = "us-east-1"
     # Generation is off unless asked for. `stub` runs the whole answer path —
     # prompt, schema parsing, grounding check, fallback — with no cloud call.
+    # Live options: `bedrock-converse` (DeepSeek/Qwen/GLM/Nova via the Converse
+    # API) or `bedrock-claude` (Anthropic via the Messages path).
     answer_provider: str = "disabled"
-    answer_model_id: str = "anthropic.claude-opus-5"
+    answer_model_id: str = "deepseek.v3.2"
+    # Optional cross-encoder rerank on the similarity fallback (needs live
+    # Bedrock invoke on the rerank model). Off by default; the section route
+    # never uses it.
+    rerank_enabled: bool = False
     metrics_enabled: bool = True
     entities_path: Path = (
         Path(__file__).resolve().parents[2]