Add Skip Straight to SQL Query Execution Option and Include REsults (#34)

* Add option to pre-run the first option in the query cache.

* Speed up existing approach for vector and cache

* Add experimentation data

* Improve write up

Author: BenConstable9

.devcontainer/devcontainer.json

@@ -9,7 +9,8 @@
         "ms-python.python",
         "ms-python.vscode-pylance",
         "ms-toolsai.jupyter",
-        "ms-azuretools.vscode-docker"
+        "ms-azuretools.vscode-docker",
+        "ms-azuretools.vscode-azurefunctions"
       ],
       "settings": {
         "editor.formatOnSave": true

deploy_ai_search/deploy.py

@@ -4,6 +4,9 @@
 from rag_documents import RagDocumentsAISearch
 from text_2_sql import Text2SqlAISearch
 from text_2_sql_query_cache import Text2SqlQueryCacheAISearch
+import logging
+
+logging.basicConfig(level=logging.INFO)
 
 
 def deploy_config(arguments: argparse.Namespace):

deploy_ai_search/text_2_sql.py

@@ -93,7 +93,8 @@ def get_index_fields(self) -> list[SearchableField]:
                 type=SearchFieldDataType.String,
                 collection=True,
                 hidden=True,
-            ),  # This is needed to enable semantic searching against the column names as complex field types are not used.
+                # This is needed to enable semantic searching against the column names as complex field types are not used.
+            ),
             SimpleField(
                 name="DateLastModified",
                 type=SearchFieldDataType.DateTimeOffset,
@@ -213,7 +214,8 @@ def get_indexer(self) -> SearchIndexer:
                     target_field_name="ColumnNames",
                 ),
                 FieldMapping(
-                    name="DateLastModified", source="/document/DateLastModified"
+                    source_field_name="/document/DateLastModified",
+                    target_field_name="DateLastModified",
                 ),
             ],
             parameters=indexer_parameters,

deploy_ai_search/text_2_sql_query_cache.py

@@ -55,12 +55,6 @@ def get_index_fields(self) -> list[SearchableField]:
             SearchableField(
                 name="Query", type=SearchFieldDataType.String, filterable=True
             ),
-            SearchField(
-                name="QueryEmbedding",
-                type=SearchFieldDataType.Collection(SearchFieldDataType.Single),
-                vector_search_dimensions=self.environment.open_ai_embedding_dimensions,
-                vector_search_profile_name=self.vector_search_profile_name,
-            ),
             ComplexField(
                 name="Schemas",
                 collection=True,

text_2_sql/.env

@@ -5,6 +5,7 @@ OpenAI__ApiKey=<openAIKey if using non managed identity>
 OpenAI__ApiVersion=<openAIApiVersion>
 Text2Sql__DatabaseEngine=<databaseEngine>
 Text2Sql__UseQueryCache=<whether to use the query cache first or not>
+Text2Sql__PreRunQueryCache=<whether to pre-run the top result from the query cache or not>
 Text2Sql__DatabaseName=<databaseName>
 Text2Sql__DatabaseConnectionString=<databaseConnectionString>
 AIService__AzureSearchOptions__Endpoint=<searchServiceEndpoint>

text_2_sql/README.md

@@ -5,7 +5,7 @@
 import os
 import json
 import logging
-from ai_search import run_ai_search_query
+from utils.ai_search import run_ai_search_query
 
 
 class AISearchPlugin:

text_2_sql/images/Known Question Response Time.png

@@ -1,11 +1,11 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT License.
 from semantic_kernel.functions import kernel_function
-import aioodbc
 from typing import Annotated
-import os
 import json
 import logging
+import os
+import aioodbc
 
 
 class PromptBasedSQLPlugin:
@@ -32,7 +32,8 @@ def load_entities(self):
             for entity_object in entities:
                 entity = entity_object["Entity"]
                 entity_object["SelectFromEntity"] = f"{self.database}.{entity}"
-                self.entities[entity.lower()] = entity_object
+                entity_name = entity_object["EntityName"].lower()
+                self.entities[entity_name] = entity_object
 
     def system_prompt(self, engine_specific_rules: str | None = None) -> str:
         """Get the schemas for the database entities and provide a system prompt for the user.
@@ -54,9 +55,14 @@ def system_prompt(self, engine_specific_rules: str | None = None) -> str:
         entity_descriptions = "\n\n        ".join(entity_descriptions)
 
         if engine_specific_rules:
-            engine_specific_rules = f"\n        The following {self.target_engine} Syntax rules must be adhered to.\n        {engine_specific_rules}"
+            engine_specific_rules = f"""\n        The following {
+                self.target_engine} Syntax rules must be adhered to.\n        {engine_specific_rules}"""
+
+        system_prompt = f"""Use the names and descriptions of {self.target_engine} entities provided in ENTITIES LIST to decide which entities to query if you need to retrieve information from the database. Use the 'GetEntitySchema()' function to get more details of the schema of the view you want to query.
+
+        Always then use the 'RunSQLQuery()' function to run the SQL query against the database. Never just return the SQL query as the answer.
 
-        system_prompt = f"""Use the names and descriptions of {self.target_engine} entities provided in ENTITIES LIST to decide which entities to query if you need to retrieve information from the database. Use the 'GetEntitySchema()' function to get more details of the schema of the view you want to query. Use the 'RunSQLQuery()' function to run the SQL query against the database.
+        Do not give the steps to the user in the response. Make sure to execute the SQL query and return the results in the response.
 
         You must always examine the provided {self.target_engine} entity descriptions to determine if they can answer the question.
 
@@ -111,7 +117,7 @@ async def get_entity_schema(
         return json.dumps({entity_name: self.entities[entity_name.lower()]})
 
     @kernel_function(
-        description="Runs an SQL query against the SQL Database to extract information.",
+        description="Runs an SQL query against the SQL Database to extract information. This function must always be used during the answer generation process. Do not just return the SQL query as the answer.",
         name="RunSQLQuery",
     )
     async def run_sql_query(