[Bug] Tabular SK analysis: multi-endpoint DeploymentNotFound 404 and Python 3.13 Optional[str] type error

[vivche]

Two related issues were identified and resolved in this release while diagnosing why tabular SK analysis was falling back to "schema context instead" when querying an Excel file (NIST SP-800-53) using a multi-endpoint model (gpt-5.1).

---

Issue 1 — Tabular SK Analysis: DeploymentNotFound 404 with Multi-Endpoint Models

Issue

When a user selected a model via the multi-endpoint model feature (e.g. a gpt-5.1 deployment hosted on a different Azure OpenAI endpoint than the default), the Semantic Kernel tabular analysis would silently fail with:

[Tabular SK Analysis] Attempt 1 synthesis failed after tool execution setup:
NotFoundError("Error code: 404 - {'error': {'code': 'DeploymentNotFound',
'message': 'The API deployment for this resource does not exist...'}}")

The analysis then fell back to returning only schema/preview context, producing degraded responses with no actual data analysis.

Root Cause

run_tabular_sk_analysis always built its AzureChatCompletion Semantic Kernel service using settings.get('azure_openai_gpt_endpoint') — the default/legacy endpoint stored in app settings.

When the main chat route resolved a multi-endpoint model, it unpacked:

gpt_client, gpt_model, gpt_provider, gpt_endpoint, gpt_auth, gpt_api_version = multi_endpoint_config

The resolved gpt_model (gpt-5.1) was passed to the tabular analysis, but gpt_endpoint and gpt_auth were not — so SK tried to call gpt-5.1 at the wrong (legacy) endpoint, which returned a 404.

Issue 2 — Tabular SK Plugin: FunctionExecutionException for Optional[str] Parameters on Python 3.13

Issue

After the endpoint fix allowed SK to successfully reach the model, tool invocations immediately failed with:

semantic_kernel.exceptions.function_exceptions.FunctionExecutionException:
Parameter sheet_name is expected to be parsed to typing.Optional[str] but is not.

TypeError: Cannot instantiate typing.Union

The same error appeared for sheet_index, source_sheet_index, and target_sheet_index parameters across multiple kernel functions.

Root Cause

Semantic Kernel's _parse_parameter (in kernel_function_from_method.py) coerces LLM-provided string arguments by calling param_type(value). When param_type is Optional[str] (which is Union[str, None]), Python 3.13 raises TypeError: Cannot instantiate typing.Union — SK cannot instantiate the Union type itself.

The affected parameters were typed as Annotated[Optional[str], "description"] in the @kernel_function decorated methods of TabularProcessingPlugin. SK successfully registered these functions but failed at invocation time when the LLM passed a value.