Added files

Chatbot.py

@@ -1,20 +1,30 @@
 from dotenv import load_dotenv
 import streamlit as st
-from utils._admin_util import invoke_rag, get_ticket_category
 import os
+from utils._graph_util import run_customer_support
+
 
 # Initialize categories in session state
 if "categories" not in st.session_state:
     st.session_state.categories = {
-        "HR Support": [],
-        "IT Support": [],
-        "Transportation Support": [],
+        "HR": [],
+        "IT": [],
+        "Transportation": [],
         "Other": []
     }
 
 def main():
     load_dotenv()
     
+    # Add detailed API key verification
+    api_key = os.getenv("OPENAI_API_KEY")
+    if not api_key:
+        st.error("❌ OpenAI API key not found! Please ensure it's set in the environment variables.")
+        st.info("To set up your API key:")
+        st.code("1. Go to Hugging Face Space settings\n2. Add OPENAI_API_KEY in Repository Secrets")
+        st.stop()
+        
+    
     # Page configuration
     st.set_page_config(
         page_title="Intelligent Customer Support Agent",
@@ -35,11 +45,6 @@ def main():
         - Other policies
         """)
         
-    # Set OpenAI API key
-    openai_api_key = os.getenv("OPENAI_API_KEY")
-    if not openai_api_key:
-        st.error("OpenAI API key not found! Please check your .env file.")
-        st.stop()
 
     # Main chat interface
     st.title("🤖 Intelligent Customer Support Agent")
@@ -55,15 +60,15 @@ def main():
         if "vector_store" not in st.session_state:
             st.error("Please load the document data first!")
             st.stop()
+    
         
-        response = invoke_rag(st.session_state.vector_store, prompt)
-        st.write(response)
-
+        response = run_customer_support(prompt)
+        st.write(response.get("response"))
         #Button to create a ticket with respective department
         button = st.button("Submit ticket?")
         
         if button:
-            category = get_ticket_category(prompt)
+            category = response.get("category")
             st.session_state.categories[category].append(prompt)
             st.success("Ticket submitted successfully!")
             # Display category (optional)

pages/Load_Documents.py

@@ -4,22 +4,6 @@ from utils._admin_util import create_embeddings, create_vector_store, read_pdf_d
 import streamlit as st
 from dotenv import load_dotenv
 
-def validate_api_key(api_key):
-    """Test if the API key is valid"""
-    try:
-       # Make a small test request to OpenAI
-        client = openai.OpenAI(api_key=api_key)
-        client.embeddings.create(input="test", model="text-embedding-ada-002")
-        return True
-    except openai.AuthenticationError:
-        st.error("❌ Invalid API key")
-        return False
-    except openai.PermissionDeniedError:
-        st.error("❌ Permission denied. Please check your API key's permissions")
-        return False
-    except Exception as e:
-        st.error(f"❌ API key validation error: {str(e)}")
-        return False
 
 def main():
     load_dotenv()
@@ -32,10 +16,6 @@ def main():
         st.code("1. Go to Hugging Face Space settings\n2. Add OPENAI_API_KEY in Repository Secrets")
         st.stop()
     
-    # Validate the API key
-    # if not validate_api_key(api_key):
-    #     st.stop()
-    
 
     st.set_page_config(page_title="Dump PDFs to QDrant - Vector Store")
     st.title("Please upload your files...📁 ")

pages/Pending_tickets.py

@@ -9,18 +9,18 @@ tabs = st.tabs(tab_titles)
 # Add content to each tab...
 with tabs[0]:
     st.header('HR Support tickets')
-    for ticket in st.session_state.categories["HR Support"]:
-        st.write(str( st.session_state.categories["HR Support"].index(ticket)+1)+" : "+ticket)
+    for ticket in st.session_state.categories["HR"]:
+        st.write(str( st.session_state.categories["HR"].index(ticket)+1)+" : "+ticket)
     
 with tabs[1]:
     st.header('IT Support tickets')
-    for ticket in st.session_state.categories['IT Support']:
-        st.write(str(st.session_state.categories['IT Support'].index(ticket)+1)+" : "+ticket)
+    for ticket in st.session_state.categories['IT']:
+        st.write(str(st.session_state.categories['IT'].index(ticket)+1)+" : "+ticket)
  
 with tabs[2]:
     st.header('Transportation Support tickets')
-    for ticket in st.session_state.categories['Transportation Support']:
-        st.write(str(st.session_state.categories['Transportation Support'].index(ticket)+1)+" : "+ticket)
+    for ticket in st.session_state.categories['Transportation']:
+        st.write(str(st.session_state.categories['Transportation'].index(ticket)+1)+" : "+ticket)
     
 with tabs[3]:
     st.header('Other tickets')

pyproject.toml

@@ -3,17 +3,16 @@ name = "midterm-streamlit"
 version = "0.1.0"
 description = "intelligent customer support chat"
 readme = "README.md"
-requires-python = ">=3.12"
+requires-python = ">=3.13"
 dependencies = [
-    "pydantic==2.10.1",
     "langchain-core==0.3.31",
-    "langchain==0.3.15",
-    "langchain-community==0.3.15",
-    "langchain-openai==0.3.1",
-    "langchain-qdrant==0.2.0",
-    "qdrant-client==1.13.2",
+    "langchain>=0.3.15",
+    "langchain-community>=0.3.15",
+    "langchain-openai>=0.3.2",
+    "langchain-qdrant>=0.2.0",
+    "qdrant-client>=1.13.2",
     "tiktoken>=0.8.0",
-    "pymupdf==1.25.2",
+    "pymupdf>=1.25.2",
     "langgraph>=0.2.67",
     "langsmith>=0.3.1",
     "openai>=1.58.1",

requirements.txt

@@ -15,4 +15,5 @@ pymupdf
 langchain-core>=0.1.0
 qdrant-client>=1.7.0
 langchain-qdrant>=0.1.0
-httpx>=0.27.2
+httpx>=0.27.2
+langgraph>=0.2.67

utils/_admin_util.py

@@ -11,31 +11,27 @@ from langchain_core.output_parsers import StrOutputParser
 from langchain_openai import ChatOpenAI
 from langchain_core.prompts import ChatPromptTemplate
 import streamlit as st
+from langchain.prompts import ChatPromptTemplate
+from dotenv import load_dotenv
+
 
 
 
 HUMAN_TEMPLATE = """
-#CONTEXT:
-{context}
 
-QUERY:
+You are a helpful assistant who answers questions based on provided context. 
+You must only use the provided context, and cannot use your own knowledge. 
+If you do not know the answer, or it's not contained in the provided context response with "I don't know"
+
+#Question:
 {query}
 
-Use the provide context to answer the provided user query. Only use the provided context to answer the query. If you do not know the answer, or it's not contained in the provided context response with "I don't know"
+#CONTEXT:
+{context}
 """
 
-# Define the system prompt for categorization
-CATEGORY_PROMPT = """You are a ticket categorization system. Categorize the following query into exactly one of these categories:
-    - HR Support: For queries about employment, benefits, leaves, workplace policies, etc.
-    - IT Support: For queries about software, hardware, network, system access, etc.
-    - Transportation Support: For queries about company transport, parking, vehicle maintenance, etc.
-    - Other: For queries that do not fit into the above categories.
-    Respond with ONLY the category name, nothing else.
-    
-    Query: {query}
-    """
-
 def check_api_key():
+    load_dotenv()
     """Verify that the API key is set and valid"""
     api_key = os.getenv("OPENAI_API_KEY")
     if not api_key:
@@ -66,10 +62,8 @@ def tiktoken_len(text):
 def split_data(text):
     try:
         text_splitter = RecursiveCharacterTextSplitter(
-            chunk_size=500,  # Increased for better context
-            chunk_overlap=50,  # Added overlap for better continuity
-            length_function=tiktoken_len,
-            separators=["\n\n", "\n", " ", ""]
+            chunk_size=1000,  # Increased for better context
+            chunk_overlap=200,  # Added overlap for better continuity
         )   
         chunks = text_splitter.split_text(text)
         if not chunks:
@@ -85,6 +79,7 @@ def create_embeddings():
         api_key = check_api_key()
         embedding_model = OpenAIEmbeddings(
             model="text-embedding-3-small",
+            openai_api_key=api_key
         )
         return embedding_model
     except Exception as e:
@@ -123,21 +118,24 @@ def create_vector_store(embedding_model, chunks):
         raise Exception(f"Error in vector store creation: {str(e)}")
 
 # create RAG
-def create_rag(vector_store):
+def create_rag():
     try:
         api_key = check_api_key()
         openai_chat_model = ChatOpenAI(
-            model="gpt-3.5-turbo",
-            openai_api_key=api_key,
-            temperature=0.7
+            model="gpt-4o-mini",
+            openai_api_key=api_key
         )
         
         chat_prompt = ChatPromptTemplate.from_messages([
             ("system", "You are a helpful assistant that answers questions based on the provided context."),
             ("human", HUMAN_TEMPLATE)
         ])
+        if 'vector_store' in st.session_state:
+            vector_store = st.session_state.vector_store        
+        else:
+            raise ValueError("Vector store not found in session state")
         
-        retriever = vector_store.as_retriever(search_kwargs={"k": 3})
+        retriever = vector_store.as_retriever(search_kwargs={"k": 5})
         
         simple_rag = (
             {"context": retriever, "query": RunnablePassthrough()}
@@ -151,37 +149,14 @@ def create_rag(vector_store):
         raise Exception(f"Error creating RAG chain: {str(e)}")
 
 # Invoke RAG
-def invoke_rag(vector_store, query):
+def invoke_rag(query):
     try:
-        rag_chain = create_rag(vector_store)
+        rag_chain = create_rag()
         response = rag_chain.invoke(query)
         return response
     except Exception as e:
         raise Exception(f"Error invoking RAG chain: {str(e)}")
 
 
-def get_ticket_category(query):
-    try:
-        api_key = check_api_key()
-        client = ChatOpenAI(
-            model="gpt-3.5-turbo",
-            openai_api_key=api_key,
-            temperature=0
-        )
-        
-        prompt = ChatPromptTemplate.from_messages([
-            ("system", CATEGORY_PROMPT)
-        ])
-        
-        chain = prompt | client | StrOutputParser()
-        category = chain.invoke({"query": query})
-        
-        category = category.strip()
-        valid_categories = ["HR Support", "IT Support", "Transportation Support", "Other"]
-        
-        return category if category in valid_categories else "Other"
-    except Exception as e:
-        st.error(f"Error in category classification: {str(e)}")
-        return "Other"  # Fallback category
 
-    
+

utils/_graph_util.py

@@ -0,0 +1,157 @@
+from turtle import st
+from typing import TypedDict, Dict
+from langgraph.graph import StateGraph, END
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.runnables.graph import MermaidDrawMethod
+from IPython.display import display , Image
+from langchain_openai import ChatOpenAI
+import os
+from dotenv import load_dotenv
+
+from utils._admin_util import create_rag
+
+class State(TypedDict):
+  query: str
+  category: str
+  sentiment: str
+  response: str
+  
+def check_api_key():
+    load_dotenv()
+    """Verify that the API key is set and valid"""
+    api_key = os.getenv("OPENAI_API_KEY")
+    print("api_key", api_key)
+    if not api_key:
+        raise ValueError("OpenAI API key not found in environment variables")
+    return api_key
+   
+api_key = check_api_key()
+
+llm = ChatOpenAI(
+            model="gpt-3.5-turbo",
+            openai_api_key=api_key,
+            temperature=0.7
+        )
+
+def rag(state: State)->State:
+    rag_chain = create_rag()
+     # Extract just the query string from the state
+    query = state["query"]
+    print("query", query)
+    response = rag_chain.invoke(query)  # Pass the string directly, not a dict
+    print("response", response)
+    return {"response": response}
+
+def categorize(state: State) -> State:
+  "HR, IT, Transportation"
+  prompt = ChatPromptTemplate.from_template(
+      "Categorize the following  query into one of these categories: "
+      "HR, IT, Transportation, Other. Query: {query}"
+  )
+  chain = prompt | llm
+  category = chain.invoke({"query": state["query"]}).content
+  return {"category": category}
+
+def analyze_sentiment(state: State) -> State:
+  prompt = ChatPromptTemplate.from_template(
+      "Analyze the sentiment of the following customer query"
+      "Response with either 'Position', 'Neutral' , or 'Negative'. Query: {query}"
+  )
+  chain = prompt | llm
+  sentiment = chain.invoke({"query": state["query"]}).content
+  return {"sentiment": sentiment}
+
+
+def handle_hr(state: State)->State:
+  prompt = ChatPromptTemplate.from_template(
+      "Provide a HR support response to the following query : {query}"
+  )
+  chain = prompt | llm
+  response = chain.invoke({"query": state["query"]}).content
+  return {"response": response}
+
+def handle_it(state: State)->State:
+  prompt = ChatPromptTemplate.from_template(
+      "Provide a IT support response to the following query : {query}"
+  )
+  chain = prompt | llm
+  response = chain.invoke({"query": state["query"]}).content
+  return {"response": response}
+
+def handle_transportation(state: State)->State:
+  prompt = ChatPromptTemplate.from_template(
+      "Provide a transportation support response to the following query : {query}"
+  )
+  chain = prompt | llm
+  response = chain.invoke({"query": state["query"]}).content
+  return {"response": response}
+
+def handle_general(state: State)->State:
+  prompt = ChatPromptTemplate.from_template(
+      "Provide a general support response to the following query : {query}"
+  )
+  chain = prompt | llm
+  response = chain.invoke({"query": state["query"]}).content
+  return {"response": response}
+
+def escalate(state: State)->State:
+  return {"response": "This query has been escalate to a human agent due to its negative sentiment"}
+
+def route_query(state: State)->State:
+  if state["sentiment"] == "Negative":
+    return "escalate"
+  elif state["category"] == "HR":
+    return "handle_hr"
+  elif state["category"] == "IT":
+    return "handle_it"
+  elif state["category"] == "Transportation":
+    return "handle_transportation"
+  else:
+    return "handle_general"
+
+def rout_to_agent(state: State)->State:
+    if "i don't know" in state["response"].lower():
+        print(state["response"])
+        print("return analyze_sentiment")
+        return "analyze_sentiment"
+    else:
+        return "END"
+
+
+def run_customer_support(query: str)->Dict[str, str]:
+    workflow = StateGraph(State)
+    workflow.add_node("categorize", categorize)
+    workflow.add_node("rag", rag) 
+    workflow.add_node("analyze_sentiment", analyze_sentiment)
+    workflow.add_node("handle_hr", handle_hr)
+    workflow.add_node("handle_it", handle_it)
+    workflow.add_node("handle_transportation", handle_transportation)
+    workflow.add_node("escalate", escalate)
+    
+    workflow.add_edge("categorize", "rag")
+    workflow.add_conditional_edges("rag", rout_to_agent, {"analyze_sentiment": "analyze_sentiment", "END": END})  
+    workflow.add_conditional_edges(
+        "analyze_sentiment",
+        route_query,
+        {
+            "handle_hr" : "handle_hr",    
+            "handle_it" :  "handle_it",
+            "handle_transportation" : "handle_transportation",
+            "escalate": "escalate"
+        }
+    )
+
+    workflow.add_edge("handle_hr", END)
+    workflow.add_edge("handle_it", END)
+    workflow.add_edge("handle_transportation", END)
+    workflow.add_edge("escalate", END)
+
+    workflow.set_entry_point("categorize")
+
+    app  = workflow.compile()
+    results = app.invoke({"query": query})
+    return {
+        "category": results.get('category', ''),  # Returns empty string if key missing
+        "sentiment": results.get('sentiment', ''), 
+        "response": results['response']
+    }