Merge pull request #687 from Joannall/master

Add knowledge generation refinement

Author: Joannall

src/Infrastructure/BotSharp.Abstraction/Knowledges/Models/GenerateKnowledge.cs

@@ -0,0 +1,19 @@
+namespace BotSharp.Abstraction.Knowledges.Models;
+
+public class GenerateKnowledge
+{
+    [JsonPropertyName("question")]
+    public string Question { get; set; } = string.Empty;
+
+    [JsonPropertyName("answer")]
+    public string Answer { get; set; } = string.Empty;
+
+    [JsonPropertyName("refined_collection")]
+    public string RefinedCollection { get; set; } = string.Empty;
+
+    [JsonPropertyName("refine_answer")]
+    public Boolean RefineAnswer { get; set; } = false;
+
+    [JsonPropertyName("existing_answer")]
+    public string ExistingAnswer { get; set; } = string.Empty;
+}

src/Plugins/BotSharp.Plugin.KnowledgeBase/BotSharp.Plugin.KnowledgeBase.csproj

@@ -22,6 +22,7 @@
     <None Remove="data\agents\01acc3e5-0af7-49e6-ad7a-a760bd12dc40\functions\memorize_knowledge.json" />
     <None Remove="data\agents\01acc3e5-0af7-49e6-ad7a-a760bd12dc40\instructions\instruction.liquid" />
     <None Remove="data\agents\01acc3e5-0af7-49e6-ad7a-a760bd12dc40\templates\knowledge.generation.liquid" />
+    <None Remove="data\agents\01acc3e5-0af7-49e6-ad7a-a760bd12dc40\templates\knowledge.generation.refine.liquid" />
     <None Remove="data\agents\6745151e-6d46-4a02-8de4-1c4f21c7da95\templates\knowledge_retrieval.fn.liquid" />
   </ItemGroup>
 
@@ -38,6 +39,9 @@
     <Content Include="data\agents\01acc3e5-0af7-49e6-ad7a-a760bd12dc40\instructions\instruction.liquid">
       <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
     </Content>
+    <Content Include="data\agents\01acc3e5-0af7-49e6-ad7a-a760bd12dc40\templates\knowledge.generation.refine.liquid">
+      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
+    </Content>
     <Content Include="data\agents\01acc3e5-0af7-49e6-ad7a-a760bd12dc40\templates\knowledge.generation.liquid">
       <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
     </Content>

src/Plugins/BotSharp.Plugin.KnowledgeBase/Functions/GenerateKnowledgeFn.cs

@@ -1,5 +1,6 @@
 using BotSharp.Abstraction.Templating;
 using BotSharp.Core.Infrastructures;
+using Microsoft.EntityFrameworkCore.Storage.ValueConversion;
 
 namespace BotSharp.Plugin.KnowledgeBase.Functions;
 
@@ -20,14 +21,23 @@ public GenerateKnowledgeFn(IServiceProvider services, KnowledgeBaseSettings sett
 
     public async Task<bool> Execute(RoleDialogModel message)
     {
-        var args = JsonSerializer.Deserialize<ExtractedKnowledge>(message.FunctionArgs ?? "{}");
+        var args = JsonSerializer.Deserialize<GenerateKnowledge>(message.FunctionArgs ?? "{}");
         var agentService = _services.GetRequiredService<IAgentService>();
         var llmAgent = await agentService.GetAgent(BuiltInAgentId.Planner);
-        var generateKnowledgePrompt = await GetGenerateKnowledgePrompt(args.Question, args.Answer);
+        var refineKnowledge = args.RefinedCollection;
+        String generateKnowledgePrompt;
+        if (args.RefineAnswer == true)
+        {
+            generateKnowledgePrompt = await GetRefineKnowledgePrompt(args.Question, args.Answer, args.ExistingAnswer);
+        }
+        else
+        {
+            generateKnowledgePrompt = await GetGenerateKnowledgePrompt(args.Question, args.Answer);
+        }
         var agent = new Agent
         {
             Id = message.CurrentAgentId ?? string.Empty,
-            Name = "sqlDriver_DictionarySearch",
+            Name = "knowledge_generator",
             Instruction = generateKnowledgePrompt,
             LlmConfig = llmAgent.LlmConfig
         };
@@ -51,6 +61,21 @@ private async Task<string> GetGenerateKnowledgePrompt(string userQuestions, stri
             { "sql_answer", sqlAnswer },
         });
     }
+    private async Task<string> GetRefineKnowledgePrompt(string userQuestion, string sqlAnswer, string existionAnswer)
+    {
+        var agentService = _services.GetRequiredService<IAgentService>();
+        var render = _services.GetRequiredService<ITemplateRender>();
+
+        var agent = await agentService.GetAgent(BuiltInAgentId.Learner);
+        var template = agent.Templates.FirstOrDefault(x => x.Name == "knowledge.generation.refine")?.Content ?? string.Empty;
+
+        return render.Render(template, new Dictionary<string, object>
+        {
+            { "user_question", userQuestion },
+            { "new_answer", sqlAnswer },
+            { "existing_answer",  existionAnswer}
+        });
+    }
     private async Task<RoleDialogModel> GetAiResponse(Agent agent)
     {
         var text = "Generate question and answer pair";

src/Plugins/BotSharp.Plugin.KnowledgeBase/data/agents/01acc3e5-0af7-49e6-ad7a-a760bd12dc40/templates/knowledge.generation.liquid

@@ -1,4 +1,4 @@
-You are a knowledge generator for knowledge base. Extract the answer in "SQL Answer" to answer the User Questions.
+You are a knowledge extractor for knowledge base. Extract the answer in "SQL Answer" to answer the User Questions.
 * Replace alias with the actual table name. Output json array only, formatting as [{"question":"string", "answer":""}]. 
 * Skip the question/answer for tmp table.
 * Don't include tmp table in the answer.

src/Plugins/BotSharp.Plugin.KnowledgeBase/data/agents/01acc3e5-0af7-49e6-ad7a-a760bd12dc40/templates/knowledge.generation.refine.liquid

@@ -0,0 +1,15 @@
+You are a knowledge extractor for knowledge base. Utilize the new answer and existing answer to generate the final integrated answer.
+Output json array only, formatting as [{"question":"", "answer":""}]. Replace the new line with \r\n.
+* Don't loss any knowledge in the existing answer.
+
+=====
+User Question:
+{{ user_question }}
+
+=====
+New Answer:
+{{ new_answer }}
+
+=====
+Existing Answer:
+{{ existing_answer }}

src/Plugins/BotSharp.Plugin.SqlDriver/data/agents/6745151e-6d46-4a02-8de4-1c4f21c7da95/functions/sql_dictionary_lookup.json

@@ -1,6 +1,6 @@
 {
   "name": "verify_dictionary_term",
-  "description": "Get id from dictionary table by keyword. Call this function only if need_lookup_dictionary is True",
+  "description": "Get id from dictionary table by keyword. Call this function only if need_lookup_dictionary is true and is_insert is false",
   "parameters": {
     "type": "object",
     "properties": {
@@ -12,6 +12,10 @@
         "type": "string",
         "description": "the reason why you need to call verify_dictionary_term"
       },
+      "is_insert": {
+        "type": "boolean",
+        "description": "if SQL statement is inserting."
+      },
       "tables": {
         "type": "array",
         "description": "all related tables",