fix: orchestrator agent citations

.github/workflows/.builds.yml

@@ -20,7 +20,7 @@ jobs:
     strategy:
       matrix:
         # Only building frontend containers to run PR based e2e tests
-        package: [frontend, api, api-ms-agent]
+        package: [frontend, api, api-ms-agent, proxy]
     timeout-minutes: 10
     steps:
       - uses: bcgov/action-builder-ghcr@v3.0.1

api-ms-agent/app/auth/service.py

@@ -64,7 +64,6 @@ async def validate_token(self, token: str) -> KeycloakUser:
                     audience=self.keycloak_client_id,
                     options={"verify_exp": True},
                 )
-                logger.info("Token decoded successfully", sub=payload.get("sub"))
             except JWTError as e:
                 logger.error("JWT verification failed", error=str(e))
                 raise HTTPException(

api-ms-agent/app/config.py

@@ -29,14 +29,14 @@ class Settings(BaseSettings):
     azure_openai_embedding_deployment: str = "text-embedding-3-large"
 
     # LLM Configuration - Low temperature for high confidence responses
-    llm_temperature: float = 0.1  # Low temperature for consistent, high-confidence responses
-    llm_max_output_tokens: int = 900  # Cap responses to control cost/token usage
+    llm_temperature: float = 0.0  # Low temperature for consistent, high-confidence responses
+    llm_max_output_tokens: int = 5000  # Cap responses to control cost/token usage
 
     # Dev UI (Agent Framework DevUI) settings
     devui_enabled: bool = True
     devui_host: str = "localhost"
     devui_port: int = 8000
-    devui_auto_open: bool = True
+    devui_auto_open: bool = False
     devui_mode: str = "developer"  # developer | user
 
     # Cosmos DB settings - for chat history, metadata, and workflow persistence
@@ -53,6 +53,10 @@ class Settings(BaseSettings):
     azure_document_intelligence_endpoint: str = ""
     azure_document_intelligence_key: str = ""  # Optional if using managed identity
 
+    # Azure Speech Services settings (for TTS)
+    azure_speech_key: str = ""
+    azure_speech_region: str = "canadacentral"
+
     # MCP base URLs for BC APIs (override defaults if needed)
     geocoder_base_url: str = ""
     orgbook_base_url: str = ""

api-ms-agent/app/devui.py

@@ -47,9 +47,7 @@ def _collect_entities() -> list[object]:
     try:
         research_agent = get_deep_research_service()._create_research_agent()  # noqa: SLF001
         entities.append(research_agent)
-        logger.info(
-            "devui_entity_added", entity=getattr(research_agent, "name", "research_agent")
-        )
+        logger.info("devui_entity_added", entity=getattr(research_agent, "name", "research_agent"))
     except Exception as exc:  # noqa: BLE001
         logger.warning("devui_research_agent_init_failed", error=str(exc))
     return entities

api-ms-agent/app/main.py

@@ -149,7 +149,7 @@ def create_app() -> FastAPI:
     app.add_middleware(AuthMiddleware)
 
     # Root endpoints (excluded from auth)
-    @app.get("/")
+    @app.get("/api")
     async def root():
         """Root endpoint - service status."""
         return {

api-ms-agent/app/middleware/security_middleware.py

@@ -17,9 +17,7 @@ async def dispatch(self, request: Request, call_next: Callable) -> Response:
         response.headers["X-Content-Type-Options"] = "nosniff"
         response.headers["X-Frame-Options"] = "DENY"
         response.headers["X-XSS-Protection"] = "1; mode=block"
-        response.headers["Strict-Transport-Security"] = (
-            "max-age=31536000; includeSubDomains"
-        )
+        response.headers["Strict-Transport-Security"] = "max-age=31536000; includeSubDomains"
         response.headers["Referrer-Policy"] = "strict-origin-when-cross-origin"
 
         # More permissive CSP for documentation pages

api-ms-agent/app/routers/__init__.py

@@ -6,13 +6,15 @@
 from app.routers.documents import router as documents_router
 from app.routers.orchestrator import router as orchestrator_router
 from app.routers.research import router as research_router
+from app.routers.speech import router as speech_router
 from app.routers.workflow_research import router as workflow_research_router
 
 api_router = APIRouter()
 api_router.include_router(chat_router, prefix="/chat", tags=["chat"])
 api_router.include_router(documents_router, prefix="/documents", tags=["documents"])
 api_router.include_router(orchestrator_router, prefix="/orchestrator", tags=["orchestrator"])
 api_router.include_router(research_router, prefix="/research", tags=["research"])
+api_router.include_router(speech_router, prefix="/speech", tags=["speech"])
 api_router.include_router(
     workflow_research_router, prefix="/workflow-research", tags=["workflow_research"]
 )

api-ms-agent/app/routers/chat.py

@@ -9,13 +9,17 @@
 
 from app.auth.dependencies import get_current_user_from_request
 from app.auth.models import KeycloakUser
+from app.logger import get_logger
 from app.services.azure_search_service import (
     AzureSearchService,
     get_azure_search_service,
 )
 from app.services.chat_agent import ChatAgentService, get_chat_agent_service
 from app.services.cosmos_db_service import CosmosDbService, get_cosmos_db_service
 from app.services.embedding_service import EmbeddingService, get_embedding_service
+from app.utils import sort_source_dicts_by_confidence
+
+logger = get_logger(__name__)
 
 router = APIRouter()
 
@@ -95,6 +99,15 @@ async def chat(
     session_id = request.session_id or str(uuid4())
     user_id = current_user.sub
 
+    logger.info(
+        "chat_request_received",
+        user_id=user_id,
+        session_id=session_id,
+        message_preview=request.message[:100],
+        has_history=request.history is not None,
+        document_id=request.document_id,
+    )
+
     # If no history provided but session_id exists, try to load from Cosmos DB
     history = None
     if request.history:
@@ -175,6 +188,7 @@ async def chat(
             message=request.message,
             history=history,
             session_id=session_id,
+            user_id=user_id,
             document_context=document_context,
         )
 
@@ -185,7 +199,9 @@ async def chat(
 
         if document_sources:
             # Use document sources from RAG search (more accurate citations)
-            for doc_src in document_sources:
+            # Sort by confidence (highest first)
+            sorted_doc_sources = sort_source_dicts_by_confidence(document_sources)
+            for doc_src in sorted_doc_sources:
                 sources.append(
                     SourceInfoResponse(
                         source_type=doc_src["source_type"],
@@ -196,6 +212,7 @@ async def chat(
                 )
         else:
             # No document context - use agent's LLM knowledge sources
+            # Sources are already sorted in the chat_agent service
             for src in result.sources:
                 sources.append(
                     SourceInfoResponse(
@@ -235,6 +252,14 @@ async def chat(
             sources=all_sources,
         )
 
+        logger.info(
+            "chat_request_completed",
+            user_id=user_id,
+            session_id=session_id,
+            source_count=len(sources),
+            has_sufficient_info=result.has_sufficient_info,
+        )
+
         return ChatResponse(
             response=result.response,
             session_id=session_id,
@@ -243,6 +268,12 @@ async def chat(
         )
 
     except Exception as e:
+        logger.error(
+            "chat_request_failed",
+            user_id=user_id,
+            session_id=session_id,
+            error=str(e),
+        )
         raise HTTPException(status_code=500, detail=f"Chat error: {str(e)}") from e
 
 

api-ms-agent/app/routers/documents.py

@@ -1,6 +1,5 @@
 """Documents router - API endpoints for document indexing and vector search."""
 
-import logging
 from datetime import datetime, UTC
 from typing import Annotated
 
@@ -9,6 +8,7 @@
 
 from app.auth.dependencies import get_current_user_from_request
 from app.auth.models import KeycloakUser
+from app.logger import get_logger
 from app.services.azure_search_service import AzureSearchService, get_azure_search_service
 from app.services.cosmos_db_service import CosmosDbService, get_cosmos_db_service
 from app.services.document_intelligence_service import (
@@ -20,7 +20,7 @@
 
 router = APIRouter()
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
 
 
 class IndexDocumentRequest(BaseModel):
@@ -135,6 +135,12 @@ async def upload_document(
         )
 
     user_id = current_user.sub if current_user else "anonymous"
+    logger.info(
+        "document_upload_requested",
+        filename=file.filename,
+        user_id=user_id,
+        content_type=file.content_type,
+    )
 
     try:
         content_bytes = await file.read()
@@ -173,6 +179,15 @@ async def upload_document(
             chunk_overlap=200,
         )
 
+        logger.info(
+            "document_upload_completed",
+            document_id=document.id,
+            filename=file.filename,
+            user_id=user_id,
+            total_chunks=len(document.chunks),
+            total_pages=total_pages,
+        )
+
         return UploadDocumentResponse(
             id=document.id,
             filename=file.filename or "unknown",
@@ -187,7 +202,12 @@ async def upload_document(
     except ValueError as e:
         raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=str(e)) from e
     except Exception as e:
-        logger.error(f"Error processing document upload: {e}")
+        logger.error(
+            "document_upload_failed",
+            filename=file.filename,
+            user_id=user_id,
+            error=str(e),
+        )
         raise HTTPException(
             status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
             detail=f"Upload error: {str(e)}",
@@ -229,6 +249,7 @@ async def list_documents(
 ) -> DocumentListResponse:
     """List all indexed documents for the current user."""
     user_id = current_user.sub
+    logger.debug("list_documents_requested", user_id=user_id, limit=limit)
 
     try:
         documents = await embedding_service.list_documents(user_id, limit)
@@ -246,6 +267,7 @@ async def list_documents(
             total=len(documents),
         )
     except Exception as e:
+        logger.error("list_documents_failed", user_id=user_id, error=str(e))
         raise HTTPException(status_code=500, detail=f"List error: {str(e)}") from e
 
 
@@ -261,6 +283,12 @@ async def index_document(
     The document is chunked and each chunk is embedded and stored in Cosmos DB.
     """
     user_id = current_user.sub
+    logger.info(
+        "document_index_requested",
+        user_id=user_id,
+        document_id=request.document_id,
+        title=request.title,
+    )
 
     try:
         document = await embedding_service.index_document(
@@ -275,6 +303,13 @@ async def index_document(
             chunk_overlap=request.chunk_overlap,
         )
 
+        logger.info(
+            "document_index_completed",
+            user_id=user_id,
+            document_id=document.id,
+            chunks_created=len(document.chunks),
+        )
+
         return IndexDocumentResponse(
             document_id=document.id,
             chunks_created=len(document.chunks),
@@ -297,6 +332,13 @@ async def search_documents(
     Perform vector similarity search across indexed documents.
     """
     user_id = current_user.sub
+    logger.info(
+        "document_search_requested",
+        user_id=user_id,
+        query_length=len(request.query),
+        document_id=request.document_id,
+        top_k=request.top_k,
+    )
 
     try:
         results = await embedding_service.search(
@@ -307,6 +349,12 @@ async def search_documents(
             min_similarity=request.min_similarity,
         )
 
+        logger.info(
+            "document_search_completed",
+            user_id=user_id,
+            results_count=len(results),
+        )
+
         return SearchResponse(
             query=request.query,
             results=[
@@ -323,6 +371,7 @@ async def search_documents(
         )
 
     except Exception as e:
+        logger.error("document_search_failed", user_id=user_id, error=str(e))
         raise HTTPException(status_code=500, detail=f"Search error: {str(e)}") from e
 
 
@@ -334,15 +383,32 @@ async def delete_document(
 ) -> dict:
     """Delete a document and all its chunks."""
     user_id = current_user.sub
+    logger.info(
+        "document_delete_requested",
+        user_id=user_id,
+        document_id=document_id,
+    )
 
     try:
         count = await embedding_service.delete_document(document_id, user_id)
+        logger.info(
+            "document_delete_completed",
+            user_id=user_id,
+            document_id=document_id,
+            chunks_deleted=count,
+        )
         return {
             "status": "deleted",
             "document_id": document_id,
             "chunks_deleted": count,
         }
     except Exception as e:
+        logger.error(
+            "document_delete_failed",
+            user_id=user_id,
+            document_id=document_id,
+            error=str(e),
+        )
         raise HTTPException(status_code=500, detail=f"Delete error: {str(e)}") from e
 
 

api-ms-agent/app/routers/orchestrator.py

@@ -92,6 +92,7 @@ async def query_orchestrator(
         result = await orchestrator.process_query(
             query=request.query,
             session_id=request.session_id,
+            user_id=current_user.sub,
         )
 
         # Map sources to the response model