Upload app_streamlit.py

app_streamlit.py

@@ -0,0 +1,245 @@
+import streamlit as st
+import os
+import requests
+from dotenv import load_dotenv  # Only needed if using a .env file
+
+# Langchain and HuggingFace
+from langchain.vectorstores import Chroma
+from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain_groq import ChatGroq
+from langchain.chains import RetrievalQA
+
+# Load the .env file (if using it)
+load_dotenv()
+groq_api_key = os.getenv("GROQ_API_KEY")
+
+# Load embeddings, model, and vector store
+@st.cache_resource  # Singleton, prevent multiple initializations
+def init_chain():
+    model_kwargs = {'trust_remote_code': True}
+    embedding = HuggingFaceEmbeddings(model_name='nomic-ai/nomic-embed-text-v1.5', model_kwargs=model_kwargs)
+    llm = ChatGroq(groq_api_key=groq_api_key, model_name="llama3-70b-8192", temperature=0.2)
+    vectordb = Chroma(persist_directory='updated_CSPCDB2', embedding_function=embedding)
+
+    # Create chain
+    chain = RetrievalQA.from_chain_type(llm=llm,
+                                        chain_type="stuff",
+                                        retriever=vectordb.as_retriever(k=5),
+                                        return_source_documents=True)
+    return chain
+
+# Streamlit app layout
+st.set_page_config(
+    page_title="CSPC Citizens Charter Conversational Agent",
+    page_icon="cspclogo.png"
+)
+
+with st.sidebar:
+    st.title('CSPCean Conversational Agent')
+    st.subheader('Ask anything CSPC Related here!')
+
+    st.markdown('''**About CSPC:**
+        History, Core Values, Mission and Vision''')
+
+    st.markdown('''**Admission & Graduation:**
+        Apply, Requirements, Process, Graduation''')
+
+    st.markdown('''**Student Services:**
+         Scholarships, Orgs, Facilities''')
+
+    st.markdown('''**Academics:**
+         Degrees, Courses, Faculty''')
+
+    st.markdown('''**Officials:**
+         President, VPs, Deans, Admin''')
+
+    st.markdown('''
+        Access the resources here:
+
+        - [CSPC Citizen’s Charter](https://cspc.edu.ph/governance/citizens-charter/)
+        - [About CSPC](https://cspc.edu.ph/about/)
+        - [College Officials](https://cspc.edu.ph/college-officials/)
+    ''')
+    st.markdown('Team XceptionNet')
+
+# Store LLM generated responses
+if "messages" not in st.session_state:
+    st.session_state.chain = init_chain()
+    st.session_state.messages = [{"role": "assistant", "content": "How may I help you today?"}]
+
+# Function for generating response using the last three conversations
+def generate_response(prompt_input):
+    # Initialize result
+    result = ''
+    
+    # Prepare conversation history: get the last 3 user and assistant messages
+    conversation_history = ""
+    recent_messages = st.session_state.messages[-3:]  # Last 3 user and assistant exchanges (each exchange is 2 messages)
+
+    for message in recent_messages:
+        conversation_history += f"{message['role']}: {message['content']}\n"
+    
+    # Append the current user prompt to the conversation history
+    conversation_history += f"user: {prompt_input}\n"
+
+    # Invoke chain with the truncated conversation history
+    res = st.session_state.chain.invoke(conversation_history)
+    
+    # Process response (as in the original code)
+    if res['result'].startswith('According to the provided context, '):
+        res['result'] = res['result'][35:]
+        res['result'] = res['result'][0].upper() + res['result'][1:]
+    elif res['result'].startswith('Based on the provided context, '):
+        res['result'] = res['result'][31:]
+        res['result'] = res['result'][0].upper() + res['result'][1:]
+    elif res['result'].startswith('According to the provided text, '):
+        res['result'] = res['result'][34:]
+        res['result'] = res['result'][0].upper() + res['result'][1:]
+    elif res['result'].startswith('According to the context, '):
+        res['result'] = res['result'][26:]
+        res['result'] = res['result'][0].upper() + res['result'][1:]
+        
+    
+    # result += res['result']
+    
+    # # Process sources
+    # result += '\n\nSources: '
+    # sources = []
+    # for source in res["source_documents"]:
+    #     sources.append(source.metadata['source'][122:-4])  # Adjust as per your source format
+
+    # sources = list(set(sources))  # Remove duplicates
+    # source_list = ", ".join(sources)
+    
+    # result += source_list
+
+    # return result, res['result'], source_list
+    # return result, res['result']
+    # def generate_response(prompt_input):
+    # # Prepare conversation history: get the last 3 user and assistant messages
+    # conversation_history = ""
+    # recent_messages = st.session_state.messages[-3:]  # Last 3 user and assistant exchanges
+
+    # for message in recent_messages:
+    #     conversation_history += f"{message['role']}: {message['content']}\n"
+    
+    # # Append the current user prompt to the conversation history
+    # conversation_history += f"user: {prompt_input}\n"
+
+    # # Invoke chain with the truncated conversation history
+    # res = st.session_state.chain.invoke(conversation_history)
+    
+    # # Process response
+    # result_text = res['result']
+    # if result_text.startswith('According to the provided context, '):
+    #     result_text = result_text[35:].capitalize()
+    # elif result_text.startswith('Based on the provided context, '):
+    #     result_text = result_text[31:].capitalize()
+    # elif result_text.startswith('According to the provided text, '):
+    #     result_text = result_text[34:].capitalize()
+    # elif result_text.startswith('According to the context, '):
+    #     result_text = result_text[26:].capitalize()
+
+    # # Extract and format sources
+    # sources = []
+    # for source in res.get("source_documents", []):  # Safeguard with .get() in case sources are missing
+    #     source_path = source.metadata.get('source', '')
+    #     formatted_source = source_path[122:-4] if source_path else "Unknown source"
+    #     sources.append(formatted_source)
+    
+    # # Remove duplicates and combine into a single string
+    # unique_sources = list(set(sources))
+    # source_list = ", ".join(unique_sources)
+    
+    # # Combine response text with sources
+    # result_text += f"\n\n**Sources:** {source_list}" if source_list else "\n\n**Sources:** None"
+    
+    # return result_text
+
+    # return res['result']
+def generate_response(prompt_input):
+    # Retrieve vector database context using ONLY the current user input
+    retriever = st.session_state.chain.retriever
+    relevant_context = retriever.get_relevant_documents(prompt_input)  # Retrieve context only for the current prompt
+    
+    # Prepare full conversation history for the LLM
+    conversation_history = ""
+    for message in st.session_state.messages:
+        conversation_history += f"{message['role']}: {message['content']}\n"
+    
+    # Append the current user prompt to the conversation history
+    conversation_history += f"user: {prompt_input}\n"
+
+    # Format the input for the chain with the retrieved context
+    formatted_input = (
+        f"Context:\n"
+        f"{' '.join([doc.page_content for doc in relevant_context])}\n\n"
+        f"Conversation:\n{conversation_history}"
+    )
+
+    # Invoke the RetrievalQA chain directly with the formatted input
+    res = st.session_state.chain.invoke({"query": formatted_input})
+    
+    # Process the response text
+    result_text = res['result']
+    if result_text.startswith('According to the provided context, '):
+        result_text = result_text[35:].capitalize()
+    elif result_text.startswith('Based on the provided context, '):
+        result_text = result_text[31:].capitalize()
+    elif result_text.startswith('According to the provided text, '):
+        result_text = result_text[34:].capitalize()
+    elif result_text.startswith('According to the context, '):
+        result_text = result_text[26:].capitalize()
+
+    # Extract and format sources (if available)
+    sources = []
+    for doc in relevant_context:
+        source_path = doc.metadata.get('source', '')
+        formatted_source = source_path[122:-4] if source_path else "Unknown source"
+        sources.append(formatted_source)
+    
+    # Remove duplicates and combine into a single string
+    unique_sources = list(set(sources))
+    source_list = ", ".join(unique_sources)
+    
+    # Combine response text with sources
+    result_text += f"\n\n**Sources:** {source_list}" if source_list else "\n\n**Sources:** None"
+    
+    return result_text
+
+
+
+# Display chat messages
+for message in st.session_state.messages:
+    with st.chat_message(message["role"]):
+        st.write(message["content"])
+
+# User-provided prompt for input box
+if prompt := st.chat_input(placeholder="Ask a question..."):
+    # Append user query to session state
+    st.session_state.messages.append({"role": "user", "content": prompt})
+    with st.chat_message("user"):
+        st.write(prompt)
+
+    # Generate and display placeholder for assistant response
+    with st.chat_message("assistant"):
+        message_placeholder = st.empty()  # Placeholder for response while it's being generated
+        with st.spinner("Generating response..."):
+            # Use conversation history when generating response
+            response = generate_response(prompt)
+        message_placeholder.markdown(response)  # Replace placeholder with actual response
+        st.session_state.messages.append({"role": "assistant", "content": response})
+
+# Clear chat history function
+def clear_chat_history():
+    # Clear chat messages (reset the assistant greeting)
+    st.session_state.messages = [{"role": "assistant", "content": "How may I help you today?"}]
+    
+    # Reinitialize the chain to clear any stored history (ensures it forgets previous user inputs)
+    st.session_state.chain = init_chain()
+
+    # Clear any additional session state that might be remembering user inquiries
+    if "recent_user_messages" in st.session_state:
+        del st.session_state["recent_user_messages"]  # Clear remembered user inputs
+
+st.sidebar.button('Clear Chat History', on_click=clear_chat_history)