Enhancements in code and readme

scenarios/Assistants/bfsi-bot-in-a-box/README.md

@@ -30,6 +30,7 @@ Contoso Financial Assistant caters to below types of queries:
 ## Installation
 
 1. Clone the repository.
+2. cd azureai-samples\scenarios\Assistants\bfsi-bot-in-a-box\src\backend
 2. Install the necessary dependencies for the backend using requirements.txt
 3. Create a .env file in backend folder and specify values of following variables
       OPEN_AI_EMBEDDING_ENDPOINT=<your_open_ai_embedding_endpoint>
@@ -50,11 +51,22 @@ Contoso Financial Assistant caters to below types of queries:
 ## Usage
 
 1. Create AI Search index for user query categorization by executing get_intent_init.py 
+python bfsi_config\tools\get_intent_init.py
 2. Start the backend server using command flask run --host 0.0.0.0 --port 5007
 3. Launch the frontend application by opening `assistant.html` file in your browser.
+cd azureai-samples\scenarios\Assistants\bfsi-bot-in-a-box\src\frontend
+python -m http.server 8000
+Browse to http://localhost:8000/assistant.html
 4. Sample conversation:
-Tell me about performance of contoso financials in 2023.
-Or
-I purchased a Washing Machine on EMI Card. 
-My EMI is 1000 Rs for 12 months.  
-I missed my last EMI. How much do i pay now?
+- Hi
+- Can you tell me more about contoso?
+- Can you create html table with all details?
+- Super, I missed my installment for washing machine last month. 
+Do I need to pay interest?
+- I have taken loan for 2 years. Every Month I pay 1000. 
+If I miss all my payments, how much will I pay finally?
+- I need detailed calculation for each month showing 
+how you arrived at the final figure.
+Can you create html table showing details for each of the 24 months?
+- Please show for all months
+- Thanks - this is super helpful!

scenarios/Assistants/bfsi-bot-in-a-box/src/backend/bfsi_config/prompts/prompt.txt

@@ -8,9 +8,10 @@ You handle only following type of queries:
 You follow below mentioned guidelines to answer user queries:
 1. Responses should be concise, simple, clear and easy to understand.
 2. Enable user to get answers in minimal iterations.
-3. If the text response is long, organize it as list of points for better readability.
+3. If the text response is long, return response as a table.
 4. Do not have non ascii characters in the response.
-5. If any table data is requested, present it in the form of a tabular chart image.
+5. For tables, use readable fonts, left aligned text, return as html text.
+6. When asked for table, do not create an image, return as html text.
 -----------------------------------------------------------------------------
 7. If the user thanks you, you revert with a summary, category & subcategory of the summary 
 8. You Identify category and sub-category using func categorize_user_query only after user thanks.

scenarios/Assistants/bfsi-bot-in-a-box/src/backend/bfsi_config/tools/get_intent_init.py

@@ -1,4 +1,5 @@
 import os
+import sys
 from dotenv import load_dotenv
 from azure.core.credentials import AzureKeyCredential
 from azure.search.documents import SearchClient
@@ -19,6 +20,8 @@
     SemanticField,
 )
 import json
+cwd = os.path.dirname(__file__)
+sys.path.append(os.path.abspath(os.path.join(cwd, '..', '..')))
 from bfsi_config.tools.open_ai_response import get_embeddings
 
 # Use this when running the code in the same directory as open_ai_response.py
@@ -87,7 +90,8 @@ def get_index(name: str) -> SearchIndex:
 
 
 def get_intent_documents() -> list:
-    file_path = "get_intent_data.json"
+
+    file_path = os.path.join(cwd, "get_intent_data.json")
     with Path(file_path).open("r") as file:
         intent_data = json.load(file)
     docs = []

scenarios/Assistants/bfsi-bot-in-a-box/src/backend/bfsi_config/tools/settings.py

@@ -10,12 +10,12 @@
 
 tools_list = [
     {"type": "code_interpreter"},
+
     # {"type": "retrieval"},
     {
         "type": "function",
         "function": {
             "name": "search_web_with_freshness_filter",
-            "module": "functions.user_functions",
             "description": """Generates response to the query using knowledgebase 
             it retrives by searching the web for information related to the input 
             query and filters results by freshness if specified.""",
@@ -34,18 +34,13 @@
                     },
                 },
                 "required": ["query"],
-            },
-            "returns": {
-                "type": "string",
-                "description": "Response to the query on the basis of web search.",
-            },
+            }
         },
     },
     {
         "type": "function",
         "function": {
             "name": "categorize_user_query",
-            "module": "functions.user_functions",
             "description": "Determines and returns the Category & Subcategory for the input user query.",
             "parameters": {
                 "type": "object",
@@ -56,11 +51,7 @@
                     }
                 },
                 "required": ["user_query"],
-            },
-            "returns": {
-                "type": "string",
-                "description": "String containing Category & Subcategory for the input query",
-            },
+            }
         },
     },
 ]

scenarios/Assistants/bfsi-bot-in-a-box/src/frontend/assistant.html

@@ -410,12 +410,18 @@
       prv_messages.push(decodedContent)
       assistantResponse = decodedContent;
 
+      let tableStartIndex = assistantResponse.indexOf('<table');
+      let tableEndIndex = assistantResponse.indexOf('</table>') + 8; // 8 is the length of '</table>'
+      if (tableStartIndex !== -1 && tableEndIndex !== -1) {
+          assistantResponse = assistantResponse.substring(tableStartIndex, tableEndIndex);
+      }
+
       var chatArea = document.getElementById('chatArea');
       var assistantDiv = document.createElement('div');
 
       assistantDiv.className = 'chat-message assistant-message';
       assistantDiv.style.flexGrow = '1'; 
-      assistantDiv.textContent = assistantResponse;
+      assistantDiv.innerHTML = assistantResponse;
 
       // add image called assistant_icon.png
       var img = document.createElement('img');