Wire the guarded conversational RAG answer layer end-to-end

apps/ai-service/config.py

@@ -0,0 +1,40 @@
+from __future__ import annotations
+
+from functools import lru_cache
+from pathlib import Path
+
+from pydantic import Field
+from pydantic_settings import BaseSettings, SettingsConfigDict
+
+
+class Settings(BaseSettings):
+    model_config = SettingsConfigDict(env_file=".env", extra="ignore")
+
+    app_name: str = "vsf-duoc-thu-ai-service"
+    qdrant_url: str = "http://localhost:6333"
+    qdrant_collection: str = "duocthu_v1"
+    qdrant_api_key: str | None = None
+    postgres_dsn: str = Field(
+        default="postgresql://duoc_thu:duoc_thu@localhost:5432/duoc_thu",
+        repr=False,
+    )
+    # Defaults to the route that is measured and needs no provider. Raising it
+    # to `cohere-v4` enables the similarity fallback and requires live Bedrock.
+    embedding_provider: str = "section-only"
+    embedding_dimensions: int = 1024
+    evidence_minimum_score: float = 0.12
+    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.
+    answer_provider: str = "disabled"
+    answer_model_id: str = "anthropic.claude-opus-5"
+    metrics_enabled: bool = True
+    entities_path: Path = (
+        Path(__file__).resolve().parents[2]
+        / "ingestion/data/verified/drug_entities.json"
+    )
+
+
+@lru_cache
+def get_settings() -> Settings:
+    return Settings()