more conversational, using objections

classes.py

@@ -3,7 +3,7 @@ class SessionState:
     do_evaluation = False
     do_opportunity_analysis = False
     do_customer_research = False
-    do_objections = True
+    do_objections = False
     add_objections_to_analysis = False   
     ask_objections = True
     do_ragas_evaluation = False

data/HSBC_background.txt

@@ -0,0 +1,47 @@
+HSBC Holdings Plc (HSBC) is a provider of banking and financial solutions. Its
+subsidiaries offer retail, business, and commercial banking; private banking, and wealth
+management services. The group’s portfolio of offerings comprises personal and
+commercial loans, deposit services, mortgages, card products, insurance solutions,
+global asset management, working capital and term loans, payment services,
+international trade facilitation, merger and acquisition advisory, transaction banking,
+capital markets, and risk management, investment management, and trusts and estate
+planning, among others. The group has a presence in Europe, Asia, the Middle East,
+North Africa, North America, and Latin America. HSBC is headquartered in London, the
+UK.
+
+From the Group Chairman
+Technology and sustainability are two of the trends transforming banking and the world around us. The opportunities from generative
+AI are among the most transformative within my working life. We are actively exploring a number of use cases, while also working to
+manage the associated risks.
+From Group Chief Executive
+Driving cost savings enables us to invest in technology, which is the fourth opportunity. The digitisation of our business continues to
+improve customer experience and increase efficiency. Using AI to help price complex structural options in our Foreign Exchange
+business has cut execution times down from hours to minutes. We have also identified hundreds of opportunities to leverage
+generative AI, and will focus our efforts on use cases with tangible benefits for the Group and our customers.
+Innovation also creates new avenues for growth. We recently launched Zing, which is our open market mobile platform focused on
+cross-border payments, initially available in the UK. It offers similar capabilities as Global Money does to our international Wealth
+and Personal Banking customers, but is targeted at non-HSBC customers and allows us to drive growth beyond our traditional
+customer footprint.
+From the Strategic Report 2023
+HSBC Technology Priority
+Digitize
+Improve customer experience and efficiency while investing in innovation
+In 2023, we made progress on our goal to become a digital-first bank, and our customers have been increasingly adopting our digital
+services. In CMB, 83% of customers were digitally active, an increase of 5 percentage points since 2022. Our net promoter score for
+onboarding wholesale international clients in the last quarter of 2023 improved by 12 points when compared with the first three
+months
+of the year. At 54%, more than half of WPB customers were mobile active, an increase of 6 percentage points from 2022.
+Furthermore, a total of 75% of WPB’s international customer accounts were opened digitally in 2023, an increase of 30 percentage
+points from 2022.
+We are also focused on building future-ready business models by investing in open-market propositions. In 2023, we announced a
+partnership with Tradeshift to launch a new embedded finance solution in the first half
+
+of 2024, which will provide payment and financial services embedded into trade, e-commerce and marketplace experiences. In
+January 2024, in the UK we launched Zing, a mobile platform enabling cross-border payments available to non-HSBC consumers.
+We are also investing in innovative technologies for the future. In 2024, we plan to both concentrate our efforts and increase our
+investment in artificial intelligence (‘Al’).
+At present, we employ Al in areas such as fraud detection and transaction monitoring. We also launched Al Markets, a digital service
+that utilises natural language processing to enrich the way investors interact with global markets. Additionally, we are in the process
+of piloting numerous generative Al use
+cases in areas like developer augmentation, creative content generation and knowledge management, and have identified
+hundreds more potential opportunities.

utils_callbacks.py

@@ -44,12 +44,18 @@ async def callback_run_scenario(action):
     session_state = cl.user_session.get("session_state", None)
     session_state.add_scenario_info(selected_opportunity) 
     get_customer_background(session_state, selected_opportunity['Customer Name'])
-    print("creating objections")
-    session_state.objections = await create_objections(session_state)
-    for obj in session_state.objections:
-        print(obj)
-    print("questions created")
-    session_state.questions = get_questions(session_state.opportunity.stage, session_state.num_questions)
+    if session_state.ask_objections:
+        print("creating objections")
+        session_state.objections = await create_objections(session_state)
+        questions = []
+        for obj in session_state.objections:
+            print(obj)
+            q = {"stage": session_state.opportunity.stage, "question": obj[3:], "ground_truth": ""}
+            questions.append(q)
+        session_state.questions = questions
+    else:
+        print("questions created")
+        session_state.questions = get_questions(session_state.opportunity.stage, session_state.num_questions)
     for q in session_state.questions:
         print(q)
     opening_message = session_state.get_opening()  

utils_chain_parameters.py

@@ -5,6 +5,7 @@ def prepare_chain_parameters(session_state, message, history):
     next_question = ""  
     ground_truth = ""
     command = ""
+    print(f"Index: {session_state.current_question_index}")
     if session_state.current_question_index == 0:
         previous_question = ""
         rep_answer = ""
@@ -18,8 +19,6 @@ def prepare_chain_parameters(session_state, message, history):
         ground_truth = session_state.questions[session_state.current_question_index - 1]["ground_truth"]
         command = """Thank the customer, offer a comment on the answer and overall performance.
             Conclude the conversation with a summary and give a farewell. 
-            If the answers were good, give a positive farewell and offer a follow up meeting. 
-            If the answers were poor, give a poor farewell.
             You can add additional comments as needed.
         """
     else:

utils_prompt.py

@@ -96,9 +96,9 @@ def get_system_template_openai_short():
     system_template = """
         You are playing a role in a conversation with a sales representative.
         Your name is in the 'Name:' section.
-        If the sales rep gets your name wrong, you can correct them, politely
         They can use your first name, full name or address you with a title and last name.
         Your name does not need to match exactly what they say.
+        Be chatty and conversational and friendly.
         Your compnay information is in the 'Company:' section.
         The sales rep's details is in the 'Sales rep:' section.
         You do not need to use the sales rep's name in your response except in the greeting.