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

apps/ai-service/pyproject.toml

@@ -1,10 +1,36 @@
 [project]
 name = "ai-service"
 version = "0.0.0"
-description = "RAG orchestration + OpenAI calls for the Duoc Thu medical chatbot"
+description = "RAG orchestration for the Duoc Thu medical chatbot"
 requires-python = ">=3.11"
-dependencies = []
+dependencies = [
+    "fastapi>=0.115,<1",
+    "httpx>=0.27,<1",
+    "psycopg[binary]>=3.2,<4",
+    "pydantic-settings>=2.6,<3",
+    "qdrant-client>=1.7,<2",
+    "uvicorn[standard]>=0.30,<1",
+]
+
+[project.optional-dependencies]
+test = ["pytest>=7.4,<9"]
+# Both optional on purpose: the service answers without a metrics stack, and
+# without a cloud generator. Neither is a precondition for a grounded answer.
+metrics = ["prometheus-client>=0.20,<1"]
+generation = ["anthropic>=0.112,<1"]
 
 [build-system]
 requires = ["setuptools>=68"]
 build-backend = "setuptools.build_meta"
+
+[tool.ruff]
+line-length = 100
+
+[tool.ruff.lint]
+select = ["F", "E9", "B", "ARG"]
+
+[tool.ruff.lint.per-file-ignores]
+# Test doubles implement the domain's Protocols. Conformance requires the full
+# signature even where a double ignores an argument, so ARG here would push
+# tests toward fakes that no longer match the interface they stand in for.
+"tests/*" = ["ARG001", "ARG002"]