Add follow up questions

app.py

@@ -176,6 +176,8 @@ def event_handling(
     tab_graphs = main_tab_components.tab_graphs
     tab_papers = main_tab_components.tab_papers
     graphs_container = main_tab_components.graph_container
+    follow_up_examples = main_tab_components.follow_up_examples
+    follow_up_examples_hidden = main_tab_components.follow_up_examples_hidden
     
     dropdown_sources = config_components.dropdown_sources
     dropdown_reports = config_components.dropdown_reports
@@ -196,15 +198,20 @@ def event_handling(
         # Event for textbox
         (textbox
             .submit(start_chat, [textbox, chatbot, search_only], [textbox, tabs, chatbot, sources_raw], queue=False, api_name=f"start_chat_{textbox.elem_id}")
-            .then(chat, [textbox, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs], concurrency_limit=8, api_name=f"chat_{textbox.elem_id}")
+            .then(chat, [textbox, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs, follow_up_examples.dataset], concurrency_limit=8, api_name=f"chat_{textbox.elem_id}")
             .then(finish_chat, None, [textbox], api_name=f"finish_chat_{textbox.elem_id}")
         )
         # Event for examples_hidden
         (examples_hidden
             .change(start_chat, [examples_hidden, chatbot, search_only], [examples_hidden, tabs, chatbot, sources_raw], queue=False, api_name=f"start_chat_{examples_hidden.elem_id}")
-            .then(chat, [examples_hidden, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs], concurrency_limit=8, api_name=f"chat_{examples_hidden.elem_id}")
+            .then(chat, [examples_hidden, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs,follow_up_examples.dataset], concurrency_limit=8, api_name=f"chat_{examples_hidden.elem_id}")
             .then(finish_chat, None, [textbox], api_name=f"finish_chat_{examples_hidden.elem_id}")
         )
+        (follow_up_examples_hidden
+            .change(start_chat, [examples_hidden, chatbot, search_only], [follow_up_examples_hidden, tabs, chatbot, sources_raw], queue=False, api_name=f"start_chat_{examples_hidden.elem_id}")
+            .then(chat, [examples_hidden, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs,follow_up_examples.dataset], concurrency_limit=8, api_name=f"chat_{examples_hidden.elem_id}")
+            .then(finish_chat, None, [textbox], api_name=f"finish_chat_{follow_up_examples_hidden.elem_id}")
+        )
     
     elif tab_name == "Beta - POC Adapt'Action":
         print("chat poc - message sent")

climateqa/chat.py

@@ -131,6 +131,7 @@ async def chat_stream(
     retrieved_contents = []
     answer_message_content = ""
     vanna_data = {}
+    follow_up_examples = gr.Dataset(samples=[])
 
     # Define processing steps
     steps_display = {
@@ -202,7 +203,12 @@ async def chat_stream(
                         sub_questions = [q["question"] + "-> relevant sources : " + str(q["sources"]) for q in event["data"]["output"]["questions_list"]]
                         history[-1].content += "Decompose question into sub-questions:\n\n - " + "\n - ".join(sub_questions)
 
-            yield history, docs_html, output_query, output_language, related_contents, graphs_html#, vanna_data
+                # Handle follow up questions
+                if event["name"] == "generate_follow_up" and event["event"] == "on_chain_end": 
+                    follow_up_examples = event["data"]["output"].get("follow_up_questions", [])
+                    follow_up_examples = gr.Dataset(samples= [ [question] for question in follow_up_examples ])
+
+            yield history, docs_html, output_query, output_language, related_contents, graphs_html, follow_up_examples#, vanna_data
 
     except Exception as e:
         print(f"Event {event} has failed")
@@ -213,4 +219,4 @@ async def chat_stream(
     # Call the function to log interaction
     log_interaction_to_azure(history, output_query, sources, docs, share_client, user_id)
 
-    yield history, docs_html, output_query, output_language, related_contents, graphs_html#, vanna_data
+    yield history, docs_html, output_query, output_language, related_contents, graphs_html, follow_up_examples#, vanna_data

climateqa/engine/chains/follow_up.py

@@ -0,0 +1,32 @@
+from typing import List
+from langchain.prompts import ChatPromptTemplate
+
+
+FOLLOW_UP_TEMPLATE = """Based on the previous question and answer, generate 2-3 relevant follow-up questions that would help explore the topic further.
+
+Previous Question: {user_input}
+Previous Answer: {answer}
+
+Generate short, concise, focused follow-up questions
+You don't need a full question as it will be reformulated later as a standalone question with the context. Eg. "Details the first point"
+"""
+
+def make_follow_up_node(llm):
+    prompt = ChatPromptTemplate.from_template(FOLLOW_UP_TEMPLATE)
+    
+    def generate_follow_up(state):
+        if not state.get("answer"):
+            return state
+            
+        response = llm.invoke(prompt.format(
+            user_input=state["user_input"],
+            answer=state["answer"]
+        ))
+        
+        # Extract questions from response
+        follow_ups = [q.strip() for q in response.content.split("\n") if q.strip()]
+        state["follow_up_questions"] = follow_ups
+        
+        return state
+        
+    return generate_follow_up

climateqa/engine/graph.py

@@ -24,6 +24,7 @@ from .chains.answer_rag import make_rag_node
 from .chains.graph_retriever import make_graph_retriever_node
 from .chains.chitchat_categorization import make_chitchat_intent_categorization_node
 from .chains.standalone_question import make_standalone_question_node
+from .chains.follow_up import make_follow_up_node  # Add this import
 
 class GraphState(TypedDict):
     """
@@ -50,6 +51,7 @@ class GraphState(TypedDict):
     recommended_content : List[Document] # OWID Graphs  # TODO merge with related_contents
     search_only : bool = False
     reports : List[str] = []
+    follow_up_questions: List[str] = []
 
 def dummy(state):
     return 
@@ -121,6 +123,11 @@ def route_retrieve_documents(state):
         return END
     return sources_to_retrieve
 
+def route_follow_up(state):
+    if state["follow_up_questions"]:
+        return "process_follow_up"
+    return END
+
 def make_id_dict(values):
     return {k:k for k in values}
 
@@ -141,6 +148,7 @@ def make_graph_agent(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_regi
     answer_rag = make_rag_node(llm, with_docs=True)
     answer_rag_no_docs = make_rag_node(llm, with_docs=False)
     chitchat_categorize_intent = make_chitchat_intent_categorization_node(llm)
+    generate_follow_up = make_follow_up_node(llm)
 
     # Define the nodes
     # workflow.add_node("set_defaults", set_defaults)
@@ -158,6 +166,8 @@ def make_graph_agent(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_regi
     workflow.add_node("retrieve_documents", retrieve_documents)
     workflow.add_node("answer_rag", answer_rag)
     workflow.add_node("answer_rag_no_docs", answer_rag_no_docs)
+    workflow.add_node("generate_follow_up", generate_follow_up)
+    # workflow.add_node("process_follow_up", standalone_question_node)
 
     # Entry point
     workflow.set_entry_point("standalone_question")
@@ -192,6 +202,12 @@ def make_graph_agent(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_regi
         make_id_dict(["retrieve_graphs", END])
     )
 
+    # workflow.add_conditional_edges(
+    #     "generate_follow_up",
+    #     route_follow_up,
+    #     make_id_dict(["process_follow_up", END])
+    # )
+
     # Define the edges
     workflow.add_edge("standalone_question", "categorize_intent")
     workflow.add_edge("translate_query", "transform_query")
@@ -200,13 +216,17 @@ def make_graph_agent(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_regi
     # workflow.add_edge("transform_query", END) # TODO remove
 
     workflow.add_edge("retrieve_graphs", END)
-    workflow.add_edge("answer_rag", END)
-    workflow.add_edge("answer_rag_no_docs", END)
+    workflow.add_edge("answer_rag", "generate_follow_up")
+    workflow.add_edge("answer_rag_no_docs", "generate_follow_up")
+    # workflow.add_edge("answer_rag", END)
+    # workflow.add_edge("answer_rag_no_docs", END)
     workflow.add_edge("answer_chitchat", "chitchat_categorize_intent")
     workflow.add_edge("retrieve_graphs_chitchat", END)
 
     # workflow.add_edge("retrieve_local_data", "answer_search")
     workflow.add_edge("retrieve_documents", "answer_search")
+    workflow.add_edge("generate_follow_up",END)
+    # workflow.add_edge("process_follow_up", "categorize_intent")
 
     # Compile
     app = workflow.compile()
@@ -246,6 +266,7 @@ def make_graph_agent_poc(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_
     answer_rag = make_rag_node(llm, with_docs=True)
     answer_rag_no_docs = make_rag_node(llm, with_docs=False)
     chitchat_categorize_intent = make_chitchat_intent_categorization_node(llm)
+    generate_follow_up = make_follow_up_node(llm)
 
     # Define the nodes
     # workflow.add_node("set_defaults", set_defaults)
@@ -265,6 +286,8 @@ def make_graph_agent_poc(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_
     workflow.add_node("retrieve_documents", retrieve_documents)
     workflow.add_node("answer_rag", answer_rag)
     workflow.add_node("answer_rag_no_docs", answer_rag_no_docs)
+    workflow.add_node("generate_follow_up", generate_follow_up)
+    workflow.add_node("process_follow_up", standalone_question_node)
 
     # Entry point
     workflow.set_entry_point("standalone_question")
@@ -299,6 +322,12 @@ def make_graph_agent_poc(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_
         make_id_dict(["retrieve_graphs", END])
     )
 
+    workflow.add_conditional_edges(
+        "generate_follow_up",
+        route_follow_up,
+        make_id_dict(["process_follow_up", END])
+    )
+
     # Define the edges
     workflow.add_edge("standalone_question", "categorize_intent")
     workflow.add_edge("translate_query", "transform_query")
@@ -307,6 +336,8 @@ def make_graph_agent_poc(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_
     # workflow.add_edge("transform_query", END) # TODO remove
 
     workflow.add_edge("retrieve_graphs", END)
+    workflow.add_edge("answer_rag", "generate_follow_up")
+    workflow.add_edge("answer_rag_no_docs", "generate_follow_up")
     workflow.add_edge("answer_rag", END)
     workflow.add_edge("answer_rag_no_docs", END)
     workflow.add_edge("answer_chitchat", "chitchat_categorize_intent")
@@ -314,10 +345,7 @@ def make_graph_agent_poc(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_
 
     workflow.add_edge("retrieve_local_data", "answer_search")
     workflow.add_edge("retrieve_documents", "answer_search")
-
-    # workflow.add_edge("transform_query", "retrieve_drias_data")
-    # workflow.add_edge("retrieve_drias_data", END)
-
+    workflow.add_edge("process_follow_up", "categorize_intent")
 
     # Compile
     app = workflow.compile()

front/tabs/chat_interface.py

@@ -54,7 +54,10 @@ def create_chat_interface(tab):
         max_height="80vh",
         height="100vh"
     )
-    
+    with gr.Row(elem_id="follow-up-examples"):
+        follow_up_examples_hidden = gr.Textbox(visible=False, elem_id="follow-up-hidden")
+        follow_up_examples = gr.Examples(examples=["sample_1","sample_2"], label="Follow up questions", inputs= [follow_up_examples_hidden], elem_id="follow-up-button", run_on_click=False)
+
     with gr.Row(elem_id="input-message"):
         
         textbox = gr.Textbox(
@@ -68,7 +71,7 @@ def create_chat_interface(tab):
         
         config_button = gr.Button("", elem_id="config-button")
     
-    return chatbot, textbox, config_button
+    return chatbot, textbox, config_button, follow_up_examples, follow_up_examples_hidden
 
 
         

front/tabs/main_tab.py

@@ -29,6 +29,8 @@ class MainTabPanel:
     tab_graphs: gr.Tab
     tab_papers: gr.Tab
     graph_container: gr.HTML
+    follow_up_examples : gr.Examples
+    follow_up_examples_hidden : gr.Textbox
 
 def cqa_tab(tab_name):
     # State variables
@@ -37,7 +39,7 @@ def cqa_tab(tab_name):
         with gr.Row(elem_id="chatbot-row"):
             # Left column - Chat interface
             with gr.Column(scale=2):
-                chatbot, textbox, config_button = create_chat_interface(tab_name)
+                chatbot, textbox, config_button, follow_up_examples, follow_up_examples_hidden = create_chat_interface(tab_name)
 
             # Right column - Content panels
             with gr.Column(scale=2, variant="panel", elem_id="right-panel"):
@@ -91,5 +93,7 @@ def cqa_tab(tab_name):
         tab_figures=tab_figures,
         tab_graphs=tab_graphs,
         tab_papers=tab_papers,
-        graph_container=graphs_container
+        graph_container=graphs_container,
+        follow_up_examples= follow_up_examples,
+        follow_up_examples_hidden = follow_up_examples_hidden
     )

style.css

@@ -115,6 +115,11 @@ main.flex.flex-1.flex-col {
     border-radius: 40px;
     padding-left: 30px;
     resize: none;
+    background-color: #f0f8ff; /* Light blue background */
+    border: 2px solid #4b8ec3; /* Blue border */
+    font-size: 16px; /* Increase font size */
+    color: #333; /* Text color */
+    box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1); /* Add shadow */
 }
 
 #input-message > div {
@@ -474,6 +479,18 @@ a {
     text-decoration: none !important;
 }
 
+/* Follow-up Examples Styles */
+#follow-up-examples {
+    height: 15vh;
+    overflow-y: auto;
+    padding: 10px 0;
+}
+
+#follow-up-button {
+    height: 100%;
+    overflow-y: auto;
+}
+
 /* Media Queries */
 /* Desktop Media Query */
 @media screen and (min-width: 1024px) {
@@ -496,6 +513,15 @@ a {
         overflow-y: scroll !important;
     }
 
+    div#chatbot-row {
+        max-height: calc(100vh - 200px) !important;
+    }
+
+    div#chatbot {
+        height: 65vh !important;
+        max-height: 65vh !important;
+    }
+
     div#chatbot-row {
         max-height: calc(100vh - 90px) !important;
     }
@@ -514,7 +540,11 @@ a {
 /* Mobile Media Query */
 @media screen and (max-width: 767px) {
     div#chatbot {
-        height: 500px !important;
+        height: 400px !important;  /* Reduced from 500px */
+    }
+
+    #follow-up-examples {
+        height: 100px;
     }
 
     #submit-button {