added clarifying structural info to ReAct prompt (still ~69% shorter than previous prompt) and shortened the RAG prompt

3 files changed, 52 insertions, 87 deletions

diff --git a/src/client/views/nodes/chatbot/agentsystem/prompts.ts b/src/client/views/nodes/chatbot/agentsystem/prompts.ts
index 9daabc35f..7000d8634 100644
--- a/src/client/views/nodes/chatbot/agentsystem/prompts.ts
+++ b/src/client/views/nodes/chatbot/agentsystem/prompts.ts
@@ -26,12 +26,33 @@ export function getReactPrompt(tools: Tool[], summaries: () => string, chatHisto
         <point>Ensure that **ALL answers follow the answer structure**: grounded text wrapped in <grounded_text> tags with corresponding citations, normal text in <normal_text> tags, and three follow-up questions at the end.</point>
     </critical_points>
 
+    <thought_structure>
+        <thought>
+            <description>
+                Always provide a thought before each action to explain why you are choosing the next step or tool. This helps clarify your reasoning for the action you will take.
+            </description>
+        </thought>
+    </thought_structure>
+
+    <action_input_structure>
+        <action_input>
+            <action_input_description>
+                Always describe what the action will do in the <action_input_description> tag. Be clear about how the tool will process the input and why it is appropriate for this stage.
+            </action_input_description>
+            <inputs>
+                <description>
+                    Provide the actual inputs for the action in the <inputs> tag. Ensure that each input is specific to the tool being used. Inputs should match the expected parameters for the tool (e.g., a search term for the website scraper, document references for RAG).
+                </description>
+            </inputs>
+        </action_input>
+    </action_input_structure>
+
     <answer_structure>
         <answer>
             <grounded_text> - All information derived from tools or user documents must be wrapped in these tags with proper citation.</grounded_text>
-            <normal_text> - Use this tag for text not derived from tools or user documents.</normal_text>
+            <normal_text> - Use this tag for text not derived from tools or user documents. It should only be for narrative-like text or extremely common knowledge information.</normal_text>
             <citations>
-                <citation> - Provide proper citations for each <grounded_text>, referencing the tool or document chunk used.</citation>
+                <citation> - Provide proper citations for each <grounded_text>, referencing the tool or document chunk used. ENSURE THAT THERE IS A CITATION WHOSE INDEX MATCHES FOR EVERY GROUNDED TEXT CITATION INDEX. </citation>
             </citations>
             <follow_up_questions> - Provide exactly three user-perspective follow-up questions.</follow_up_questions>
             <loop_summary> - Summarize the actions and tools used in the conversation.</loop_summary>
@@ -41,7 +62,8 @@ export function getReactPrompt(tools: Tool[], summaries: () => string, chatHisto
     <grounded_text_guidelines>
         <step>**Wrap ALL tool-based information** in <grounded_text> tags and provide citations.</step>
         <step>Use separate <grounded_text> tags for distinct information or when switching to a different tool or document.</step>
-        <step>Ensure that **EVERY** <grounded_text> tag includes a citation index referencing the source of the information.</step>
+        <step>Ensure that **EVERY** <grounded_text> tag includes a citation index aligned with a citation that you provide that references the source of the information.</step>
+        <step>There should be a one-to-one relationship between <grounded_text> tags and citations.</step>
         <step>Over-citing is discouraged—only cite the information that is directly relevant to the user's query.</step>
     </grounded_text_guidelines>
 
diff --git a/src/client/views/nodes/chatbot/chatboxcomponents/ChatBox.tsx b/src/client/views/nodes/chatbot/chatboxcomponents/ChatBox.tsx
index 6dc691798..28bfbeae3 100644
--- a/src/client/views/nodes/chatbot/chatboxcomponents/ChatBox.tsx
+++ b/src/client/views/nodes/chatbot/chatboxcomponents/ChatBox.tsx
@@ -51,8 +51,7 @@ export class ChatBox extends ViewBoxAnnotatableComponent<FieldViewProps>() {
     private vectorstore_id: string;
     private vectorstore: Vectorstore;
     private agent: Agent;
-    private _oldWheel: HTMLDivElement | null = null;
-    private messagesRef: React.RefObject;
+    private messagesRef: React.RefObject<HTMLDivElement>;
 
     /**
      * Static method that returns the layout string for the field.
@@ -246,12 +245,12 @@ export class ChatBox extends ViewBoxAnnotatableComponent<FieldViewProps>() {
      * @param event The form submission event.
      */
     @action
-    askGPT = async (event: React.FormEvent): Promise => {
+    askGPT = async (event: React.FormEvent): Promise<any> => {
         event.preventDefault();
         this.inputValue = '';
 
         // Extract the user's message
-        const textInput = event.currentTarget.elements.namedItem('messageInput') as HTMLInputElement;
+        const textInput = (event.currentTarget as HTMLFormElement).elements.namedItem('messageInput') as HTMLInputElement;
         const trimmedText = textInput.value.trim();
 
         if (trimmedText) {
diff --git a/src/client/views/nodes/chatbot/tools/RAGTool.ts b/src/client/views/nodes/chatbot/tools/RAGTool.ts
index c24306dcd..f4b7b42ea 100644
--- a/src/client/views/nodes/chatbot/tools/RAGTool.ts
+++ b/src/client/views/nodes/chatbot/tools/RAGTool.ts
@@ -14,114 +14,58 @@ export class RAGTool extends BaseTool {
             {
                 hypothetical_document_chunk: {
                     type: 'string',
-                    description:
-                        "Detailed version of the prompt that is effectively a hypothetical document chunk that would be ideal to embed and compare to the vectors of real document chunks to fetch the most relevant document chunks to answer the user's query",
+                    description: "A detailed prompt representing an ideal chunk to embed and compare against document vectors to retrieve the most relevant content for answering the user's query.",
                     required: 'true',
                 },
             },
             `
-            Your task is to provide a comprehensive response to the user's prompt based on the given chunks and chat history. Follow these structural guidelines meticulously:
+            When using the RAG tool, the structure must adhere to the format described in the ReAct prompt. Below are additional guidelines specifically for RAG-based responses:
 
-            1. Overall Structure:
-            <answer>
-                [Main content with grounded_text tags interspersed with normal plain text (information that is not derived from chunks' information)]
-                <citations>
-                [Individual citation tags]
-                </citations>
-                <follow_up_questions>
-                [Three question tags]
-                </follow_up_questions>
-            </answer>
-
-            2. Grounded Text Tag Structure:
-            - Basic format: 
-                <grounded_text citation_index="[citation index number(s)]">
-                [Your generated text based on information from a subset of a chunk (a citation's direct text)]
-                </grounded_text>
+            1. **Grounded Text Guidelines**:
+                - Each <grounded_text> tag must correspond to exactly one citation, ensuring a one-to-one relationship.
+                - Always cite a **subset** of the chunk, never the full text. The citation should be as short as possible while providing the relevant information (typically one to two sentences).
+                - Do not paraphrase the chunk text in the citation; use the original subset directly from the chunk.
+                - If multiple citations are needed for different sections of the response, create new <grounded_text> tags for each.
 
-            3. Citation Tag Structure:
-            <citation index="[unique number]" chunk_id="[UUID v4]" type="[text/image/table]">
-                [For text: relevant subset of original chunk]
-                [For image/table: leave empty]
-            </citation>
+            2. **Citation Guidelines**:
+                - The citation must include only the relevant excerpt from the chunk being referenced.
+                - Use unique citation indices and reference the chunk_id for the source of the information.
+                - For text chunks, the citation content must reflect the **exact subset** of the original chunk that is relevant to the grounded_text tag.
 
-            4. Detailed Grounded Text Guidelines:
-                a. Wrap all information derived from chunks in grounded_text tags.
-                b. DO NOT PUT ANYTHING THAT IS NOT DIRECTLY DERIVED FROM INFORMATION FROM CHUNKS (EITHER IMAGE, TABLE, OR TEXT) IN GROUNDED_TEXT TAGS.
-                c. Use a single grounded_text tag for suquential and closely related information that references the same citation. If other citations' information are used sequentially, create new grounded_text tags.
-                d. Ensure every grounded_text tag has up to a few corresponding citations (should not be more than 3 and only 1 is fine). Multiple citation indices should be separated by commas.
-                e. Grounded text can be as short as a few words or as long as several sentences.
-                f. Avoid overlapping or nesting grounded_text tags; instead, use sequential tags.
-
-            5. Detailed Citation Guidelines:
-                a. Create a unique citation for each distinct piece of information from the chunks that is used to support grounded_text.
-                b. ALL TEXT CITATIONS must have direct text in its element content (e.g. <citation ...>DIRECT TEXT HERE</citation>) that is a relevant SUBSET of the original text chunk that is being cited specifically.
-                c. DO NOT paraphrase or summarize the text; use the original text as much as possible.
-                d. DO NOT USE THE FULL TEXT CHUNK as the citation content; only use the relevant subset of the text that the grounded_text is base. AS SHORT AS POSSIBLE WHILE PROVIDING INFORMATION (ONE TO TWO SENTENCES USUALLY)!
-                e. Ensure each citation has a unique index number.
-                f. Specify the correct type: "text", "image", or "table".
-                g. For text chunks, the content of the citation should ALWAYS have the relevant subset of the original text that the grounded_text is based on.
-                h. For image/table chunks, leave the citation content empty.
-                i. One citation can be used for multiple grounded_text tags if they are based on the same chunk information.
-                j. !!!DO NOT OVERCITE - only include citations for information that is directly relevant to the grounded_text.
-
-            6. Structural Integrity Checks:
-                a. Ensure all opening tags have corresponding closing tags.
-                b. Verify that all grounded_text tags have valid citation_index attributes (they should be equal to the associated citation(s) index field—not their chunk_id field).
-                c. Check that all cited indices in grounded_text tags have corresponding citations.
-
-            Example of grounded_text usage:
+            **Example**:
 
             <answer>
-                <grounded_text citation_index="1,2">
-                Artificial Intelligence (AI) is revolutionizing various sectors, with healthcare experiencing significant transformations in areas such as diagnosis and treatment planning.
-                </grounded_text>
-                <grounded_text citation_index="2,3,4">
-                In the field of medical diagnosis, AI has shown remarkable capabilities, particularly in radiology. For instance, AI systems have drastically improved mammogram analysis, achieving 99% accuracy at a rate 30 times faster than human radiologists.
+                <grounded_text citation_index="1">
+                    Artificial Intelligence is revolutionizing various sectors, with healthcare seeing transformations in diagnosis and treatment planning.
                 </grounded_text>
-                <grounded_text citation_index="4">
-                This advancement not only enhances the efficiency of healthcare systems but also significantly reduces the occurrence of false positives, leading to fewer unnecessary biopsies and reduced patient stress.
+                <grounded_text citation_index="2">
+                    Based on recent data, AI has drastically improved mammogram analysis, achieving 99% accuracy at a rate 30 times faster than human radiologists.
                 </grounded_text>
-            
-                <grounded_text citation_index="5,6">
-                Beyond diagnosis, AI is playing a crucial role in drug discovery and development. By analyzing vast amounts of genetic and molecular data, AI algorithms can identify potential drug candidates much faster than traditional methods.
-                </grounded_text>
-                <grounded_text citation_index="6">
-                    This could potentially reduce the time and cost of bringing new medications to market, especially for rare diseases that have historically received less attention due to limited market potential.
-                </grounded_text>
-
-                [... rest of the content ...]
 
                 <citations>
-                    <citation index="1" chunk_id="123e4567-e89b-12d3-a456-426614174000" type="text">Artificial Intelligence is revolutionizing various industries, with healthcare being one of the most profoundly affected sectors.</citation>
-                    <citation index="2" chunk_id="123e4567-e89b-12d3-a456-426614174001" type="text">AI has shown particular promise in the field of radiology, enhancing the accuracy and speed of image analysis.</citation>
-                    <citation index="3" chunk_id="123e4567-e89b-12d3-a456-426614174002" type="text">According to recent studies, AI systems have achieved 99% accuracy in mammogram analysis, performing the task 30 times faster than human radiologists.</citation>
-                    <citation index="4" chunk_id="123e4567-e89b-12d3-a456-426614174003" type="text">The improvement in mammogram accuracy has led to a significant reduction in false positives, decreasing the need for unnecessary biopsies and reducing patient anxiety.</citation>
-                    <citation index="5" chunk_id="123e4567-e89b-12d3-a456-426614174004" type="text">AI is accelerating the drug discovery process by analyzing complex molecular and genetic data to identify potential drug candidates.</citation>
-                    <citation index="6" chunk_id="123e4567-e89b-12d3-a456-426614174005" type="text">The use of AI in drug discovery could significantly reduce the time and cost associated with bringing new medications to market, particularly for rare diseases.</citation>
+                    <citation index="1" chunk_id="abc123" type="text">Artificial Intelligence is revolutionizing various industries, especially in healthcare.</citation>
+                    <citation index="2" chunk_id="abc124" type="table"></citation>
                 </citations>
 
                 <follow_up_questions>
-                    <question>How might AI-driven personalized medicine impact the cost and accessibility of healthcare in the future?</question>
-                    <question>What measures can be taken to ensure that AI systems in healthcare are free from biases and equally effective for diverse populations?</question>
-                    <question>How could the role of healthcare professionals evolve as AI becomes more integrated into medical practices?</question>
+                    <question>How can AI enhance patient outcomes in fields outside radiology?</question>
+                    <question>What are the challenges in implementing AI systems across different hospitals?</question>
+                    <question>How might AI-driven advancements impact healthcare costs?</question>
                 </follow_up_questions>
             </answer>
             `,
 
-            `Performs a RAG (Retrieval-Augmented Generation) search on user documents and returns a 
-            set of document chunks (either images or text) that can be used to provide a grounded response based on 
-            user documents`
+            `Performs a RAG (Retrieval-Augmented Generation) search on user documents and returns a set of document chunks (text or images) to provide a grounded response based on user documents.`
         );
     }
 
-    async execute(args: { hypothetical_document_chunk: string }): Promise {
+    async execute(args: { hypothetical_document_chunk: string }): Promise<any> {
         const relevantChunks = await this.vectorstore.retrieve(args.hypothetical_document_chunk);
         const formatted_chunks = await this.getFormattedChunks(relevantChunks);
         return formatted_chunks;
     }
 
-    async getFormattedChunks(relevantChunks: RAGChunk[]): Promise {
+    async getFormattedChunks(relevantChunks: RAGChunk[]): Promise<any> {
         try {
             const { formattedChunks } = await Networking.PostToServer('/formatChunks', { relevantChunks });