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

apps/ai-service/main.py

@@ -0,0 +1,55 @@
+from __future__ import annotations
+
+from typing import Any
+
+from fastapi import FastAPI, Response
+
+from adapters.postgres import PostgresTraceRepository
+from bootstrap import build_runtime
+from config import Settings, get_settings
+from rag.answer import GroundedAnswerService
+from routers.rag import router as rag_router
+
+
+def create_app(
+    *,
+    settings: Settings | None = None,
+    answer_service: GroundedAnswerService | None = None,
+    conversational: Any | None = None,
+    trace_writer: PostgresTraceRepository | None = None,
+    metrics: Any | None = None,
+) -> FastAPI:
+    configured = settings or get_settings()
+    app = FastAPI(title=configured.app_name, version="0.1.0")
+    app.state.answer_service = answer_service
+    app.state.conversational = conversational
+    app.state.trace_writer = trace_writer
+    app.state.metrics = metrics
+
+    @app.get("/health")
+    def health() -> dict[str, str]:
+        return {"status": "ok"}
+
+    @app.get("/metrics")
+    def prometheus_metrics() -> Response:
+        exporter = getattr(app.state, "metrics", None)
+        if exporter is None or not hasattr(exporter, "render"):
+            # 404 rather than an empty 200: a scrape that silently succeeds
+            # with no samples looks identical to a service answering nothing.
+            return Response(status_code=404)
+        body, content_type = exporter.render()
+        return Response(content=body, media_type=content_type)
+
+    app.include_router(rag_router)
+    return app
+
+
+_settings = get_settings()
+_answer_service, _conversational, _trace_writer, _metrics = build_runtime(_settings)
+app = create_app(
+    settings=_settings,
+    answer_service=_answer_service,
+    conversational=_conversational,
+    trace_writer=_trace_writer,
+    metrics=_metrics,
+)