Unreverse prompts and introduce relevance option (not hardcoded) (#28)

Motivation and Context (Why the change? What's the scenario?)

    Prompts got shuffled up in all the code moves.
    Customers changing relevancy setting as they experiment with sandbox

High level description (Approach, Design)

    Unreverse prompts
    Make relevancy a configurable option (not hardcoded): MinSchemaRelevance

.github/_typos.toml

@@ -11,6 +11,8 @@ extend-exclude = [
     "package-lock.json",
     "*.bicep",
     "*.sql",
+    "vocab.bpe",
+    "encoder.json"
 ]
 
 [default.extend-words]

README.md

@@ -261,7 +261,7 @@ running the service locally with OpenAPI enabled.
 3. [Using the Semantic Memory web service](examples/002-dotnet-WebClient)
 4. [How to upload files from command line with curl](examples/003-curl-calling-webservice)
 5. [Processing files with custom steps](examples/004-dotnet-ServerlessCustomPipeline)
-6. [Using a custom pipeline handler with serveless memory class](examples/005-dotnet-InProcessMemoryWithCustomHandler)
+6. [Using a custom pipeline handler with serverless memory class](examples/005-dotnet-InProcessMemoryWithCustomHandler)
 6. [Writing a custom async pipeline handler](examples/006-dotnet-CustomHandlerAsAService)
 
 ## Tools

dotnet/Service/README.md

@@ -72,7 +72,7 @@ The service depends on three main components:
 
 
 * **Embedding generator**: all the documents uploaded are automatically
-  partioned (aka "chunked") and indexed for vector search, generating
+  partitioned (aka "chunked") and indexed for vector search, generating
   several embedding vectors for each file. We recommend using
   [OpenAI ADA v2](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings)
   model, though you can easily plug in any embedding generator if needed.

examples/200-dotnet-nl2sql/nl2sql.config/nl2sql/generatequery/skprompt.txt

@@ -1,11 +1,59 @@
-If the requested OBJECTIVE can be answered by querying a database with tables described in SCHEMA, ANSWER: YES.
-Otherwise ANSWER: NO.
+Generate a SQL SELECT query that is compatible with {{$data_platform}} and achieves the OBJECTIVE exclusively using only the tables and views described in "SCHEMA:".
 
-Do not answer with any other word than YES or NO.
+Only generate SQL if the OBJECTIVE can be answered by querying a database with tables described in SCHEMA.
+
+Do not include any explanations, only provide valid SQL.
+
+[BEGIN EXAMPLE]
+
+SCHEMA:
+description: historical record of concerts, stadiums and singers
+tables:
+  - stadium: 
+    columns:
+      *: all columns
+      Stadium_ID: 
+      Location: 
+      Name: 
+      Capacity: 
+      Highest: 
+      Lowest: 
+      Average: 
+  - singer: 
+    columns:
+      *: all columns
+      Singer_ID: 
+      Name: 
+      Country: 
+      Song_Name: 
+      Song_release_year: 
+      Age: 
+      Is_male: 
+  - concert: 
+    columns:
+      *: all columns
+      concert_ID: 
+      concert_Name: 
+      Theme: 
+      Stadium_ID: 
+      Year: 
+  - singer_in_concert: 
+    columns:
+      *: all columns
+      concert_ID: 
+      Singer_ID: 
+references:
+  concert.Stadium_ID: stadium.Stadium_ID
+  singer_in_concert.concert_ID: concert.concert_ID
+  singer_in_concert.Singer_ID: singer.Singer_ID
+
+OBJECTIVE: "How many heads of the departments are older than 56 ?"
+SQL: select count(*) department_head_count from head where age > 56
+
+[END EXAMPLE]
 
 SCHEMA:
 {{$data_schema}}
 
 OBJECTIVE: {{$data_objective}}
-
-ANSWER: Let's think step by step.
+SQL: Let's think step by step.

examples/200-dotnet-nl2sql/nl2sql.config/nl2sql/isquery/skprompt.txt

@@ -1,59 +1,11 @@
-Generate a SQL SELECT query that is compatible with {{$data_platform}} and achieves the OBJECTIVE exclusively using only the tables and views described in "SCHEMA:".
+If the requested OBJECTIVE can be answered by querying a database with tables described in SCHEMA, ANSWER: YES.
+Otherwise ANSWER: NO.
 
-Only generate SQL if the OBJECTIVE can be answered by querying a database with tables described in SCHEMA.
-
-Do not include any explanations, only provide valid SQL.
-
-[BEGIN EXAMPLE]
-
-SCHEMA:
-description: historical record of concerts, stadiums and singers
-tables:
-  - stadium: 
-    columns:
-      *: all columns
-      Stadium_ID: 
-      Location: 
-      Name: 
-      Capacity: 
-      Highest: 
-      Lowest: 
-      Average: 
-  - singer: 
-    columns:
-      *: all columns
-      Singer_ID: 
-      Name: 
-      Country: 
-      Song_Name: 
-      Song_release_year: 
-      Age: 
-      Is_male: 
-  - concert: 
-    columns:
-      *: all columns
-      concert_ID: 
-      concert_Name: 
-      Theme: 
-      Stadium_ID: 
-      Year: 
-  - singer_in_concert: 
-    columns:
-      *: all columns
-      concert_ID: 
-      Singer_ID: 
-references:
-  concert.Stadium_ID: stadium.Stadium_ID
-  singer_in_concert.concert_ID: concert.concert_ID
-  singer_in_concert.Singer_ID: singer.Singer_ID
-
-OBJECTIVE: "How many heads of the departments are older than 56 ?"
-SQL: select count(*) department_head_count from head where age > 56
-
-[END EXAMPLE]
+Do not answer with any other word than YES or NO.
 
 SCHEMA:
 {{$data_schema}}
 
 OBJECTIVE: {{$data_objective}}
-SQL: Let's think step by step.
+
+ANSWER: Let's think step by step.

examples/200-dotnet-nl2sql/nl2sql.console/Nl2SqlConsole.cs

@@ -7,6 +7,7 @@
 using System.Linq;
 using System.Threading;
 using System.Threading.Tasks;
+using Microsoft.Extensions.Configuration;
 using Microsoft.Extensions.Hosting;
 using Microsoft.Extensions.Logging;
 using Microsoft.SemanticKernel;
@@ -38,13 +39,16 @@ internal sealed class Nl2SqlConsole : BackgroundService
 
     public Nl2SqlConsole(
         IKernel kernel,
+        IConfiguration config,
         SqlConnectionProvider sqlProvider,
         ILogger<Nl2SqlConsole> logger)
     {
+        var minRelevance = config.GetValue<double>("MinSchemaRelevance", SqlQueryGenerator.DefaultMinRelevance);
+
         this._kernel = kernel;
         this._sqlProvider = sqlProvider;
         this._logger = logger;
-        this._queryGenerator = new SqlQueryGenerator(this._kernel, Repo.RootConfigFolder);
+        this._queryGenerator = new SqlQueryGenerator(this._kernel, Repo.RootConfigFolder, minRelevance);
     }
 
     protected override async Task ExecuteAsync(CancellationToken stoppingToken)

examples/200-dotnet-nl2sql/nl2sql.console/appsettings.json

@@ -1,4 +1,7 @@
 {
+  // Semantic relevancy threshold for selecting schema
+  "MinSchemaRelevance": 0.7,
+  // Logging options
   "Logging": {
     "LogLevel": {
       "Default": "Trace"

examples/200-dotnet-nl2sql/nl2sql.library/SqlQueryGenerator.cs

@@ -18,6 +18,8 @@ namespace SemanticKernel.Data.Nl2Sql.Library;
 /// </summary>
 public sealed class SqlQueryGenerator
 {
+    public const double DefaultMinRelevance = 0.7D;
+
     public const string ContextParamObjective = "data_objective";
     public const string ContextParamSchema = "data_schema";
     public const string ContextParamSchemaId = "data_schema_id";
@@ -35,7 +37,7 @@ public sealed class SqlQueryGenerator
     private readonly ISKFunction _promptGenerator;
     private readonly ISemanticTextMemory _memory;
 
-    public SqlQueryGenerator(IKernel kernel, string rootSkillFolder)
+    public SqlQueryGenerator(IKernel kernel, string rootSkillFolder, double minRelevanceScore = DefaultMinRelevance)
     {
         var functions = kernel.ImportSemanticSkillFromDirectory(rootSkillFolder, SkillName);
         this._promptEval = functions["isquery"];