Update app.py

app.py

@@ -9,7 +9,7 @@ from langchain_community.document_loaders import (
     Docx2txtLoader,
     UnstructuredExcelLoader,
     JSONLoader,
-    UnstructuredFileLoader  # Generic loader, good for tables
+    UnstructuredFileLoader # Generic loader, good for tables
 )
 from langchain_text_splitters import RecursiveCharacterTextSplitter
 from langchain.embeddings import HuggingFaceEmbeddings
@@ -21,47 +21,10 @@ from langchain.schema.runnable import RunnablePassthrough
 from langchain.schema.output_parser import StrOutputParser
 
 # --- Configuration ---
-# --- Moved groq_api_key here ---
-load_dotenv()
-groq_api_key = os.getenv("GROQ_API_KEY")
-#groq_api_key = ""
 DOCS_DIR = "docs"
+# Using a local sentence transformer model for embeddings
 EMBEDDING_MODEL_NAME = "sentence-transformers/all-MiniLM-L6-v2"
-CACHE_DIR = ".streamlit_cache"
-GENERAL_QA_PROMPT = """
-You are an AI assistant for our internal knowledge base.
-Your goal is to provide accurate and concise answers based ONLY on the provided context.
-Do not make up information. If the answer is not found in the context, state that clearly.
-Ensure your answers are directly supported by the text.
-Accuracy is paramount.
-
-Context:
-{context}
-
-Question: {question}
-
-Answer:
-"""
-ORDER_STATUS_PROMPT = """
-You are an AI assistant helping with customer order inquiries.
-Based ONLY on the following retrieved information from our order system and policies:
-{context}
-
-The customer's query is: {question}
-
-Please perform the following steps:
-1. Carefully analyze the context for any order details (Order ID, Customer Name, Status, Items, Dates, etc.).
-2. If an order matching the query (or related to a name in the query) is found in the context:
-    - Address the customer by their name if available in the order details (e.g., "Hello [Customer Name],").
-    - Provide ALL available information about their order, including Order ID, status, items, dates, and any other relevant details found in the context.
-    - Be comprehensive and clear.
-3. If no specific order details are found in the context that match the query, politely state that you couldn't find the specific order information in the provided documents and suggest they contact support for further assistance.
-4. Do NOT invent or infer any information not explicitly present in the context.
-
-Answer:
-"""
-
-
+CACHE_DIR = ".streamlit_cache" # For potential disk-based caching if needed beyond Streamlit's default
 
 # Create docs and cache directory if they don't exist
 if not os.path.exists(DOCS_DIR):
@@ -92,7 +55,8 @@ def get_loader(file_path):
         return None
 
 # --- Caching Functions ---
-@st.cache_resource(show_spinner=False)  # Disable spinner during initial load
+
+@st.cache_resource(show_spinner="Loading and Processing Documents...")
 def load_and_process_documents(docs_path: str):
     """
     Loads documents from the specified path, processes them, and splits into chunks.
@@ -110,7 +74,7 @@ def load_and_process_documents(docs_path: str):
 
     for file_path in doc_files:
         try:
-            st.write(f"Processing: {os.path.basename(file_path)}...")  # Show progress
+            st.write(f"Processing: {os.path.basename(file_path)}...") # Show progress
             loader = get_loader(file_path)
             if loader:
                 loaded_docs = loader.load()
@@ -136,7 +100,7 @@ def load_and_process_documents(docs_path: str):
     st.success(f"Successfully loaded and processed {len(doc_files)} documents into {len(chunked_documents)} chunks.")
     return chunked_documents
 
-@st.cache_resource(show_spinner=False)  # Disable spinner during initial load
+@st.cache_resource(show_spinner="Creating Vector Store (Embeddings)...")
 def create_vector_store(_documents, _embedding_model_name: str):
     """Creates a FAISS vector store from the given documents and embedding model."""
     if not _documents:
@@ -149,12 +113,9 @@ def create_vector_store(_documents, _embedding_model_name: str):
         return vector_store
     except Exception as e:
         st.error(f"Error creating vector store: {e}")
-        # Return an empty FAISS instance instead of None to prevent the AttributeError.
-        embeddings = HuggingFaceEmbeddings(model_name=_embedding_model_name) # Initialize embeddings
-        vector_store = FAISS.from_documents([], embeddings)  # Changed from None to FAISS.from_documents
-        return vector_store
+        return None
 
-@st.cache_resource(show_spinner=False)  # Disable spinner during initial load
+@st.cache_resource(show_spinner="Initializing LLM...")
 def get_llm(api_key: str, model_name: str = "llama3-8b-8192"): # UPDATED MODEL
     """Initializes the Groq LLM."""
     if not api_key:
@@ -166,7 +127,7 @@ def get_llm(api_key: str, model_name: str = "llama3-8b-8192"): # UPDATED MODEL
         # "llama3-70b-8192" (more powerful, potentially slower)
         # "gemma-7b-it"
         llm = ChatGroq(temperature=0, groq_api_key=api_key, model_name=model_name)
-        st.sidebar.info(f"LLM Initialized: {model_name}")  # Add info about which model is used
+        st.sidebar.info(f"LLM Initialized: {model_name}") # Add info about which model is used
         return llm
     except Exception as e:
         st.error(f"Error initializing Groq LLM: {e}")
@@ -186,7 +147,8 @@ def get_rag_chain(llm, retriever, prompt_template):
 
 # --- Main Application Logic ---
 def main():
-
+    load_dotenv()
+    groq_api_key = os.getenv("GROQ_API_KEY")
 
     # --- UI Setup ---
     st.set_page_config(page_title="Internal Knowledge Base AI", layout="wide", initial_sidebar_state="expanded")
@@ -210,101 +172,124 @@ def main():
     </style>
     """, unsafe_allow_html=True)
 
-    st.title("📚 Internal Knowledge Base AI 💡")
+    st.title("📚 Internal Knowledge Base AI �")
 
     st.sidebar.header("System Status")
     status_placeholder = st.sidebar.empty()
     status_placeholder.info("Initializing...")
 
+
     if not groq_api_key:
         status_placeholder.error("GROQ API Key not configured. Application cannot start.")
         st.stop()
 
-    # --- Initialize session state ---
-    if "app_initialized" not in st.session_state:
-        st.session_state.app_initialized = False
-
-    # --- Start Button ---
-    if not st.session_state.app_initialized:
-        if st.button("Start"):  # Create a start button
-            st.session_state.app_initialized = True # set the session state to true
-            st.rerun()  # Rerun the app to trigger the knowledge base loading
-
-    # --- Knowledge Base Loading and LLM Initialization ---
-    if st.session_state.app_initialized: # only run if the app has been initialized
-        with st.spinner("Knowledge Base is loading... Please wait."):
-            start_time = time.time()
-            processed_documents = load_and_process_documents(DOCS_DIR)
-            if not processed_documents:
-                status_placeholder.error("Failed to load or process documents. Check logs and `docs` folder.")
-                st.stop()
-
-            vector_store = create_vector_store(processed_documents, EMBEDDING_MODEL_NAME)
-            if not vector_store:
-                status_placeholder.error("Failed to create vector store. Application cannot proceed.")
-                st.stop()
-
-            # Pass the selected model to get_llm
-            llm = get_llm(groq_api_key, model_name="llama3-8b-8192")  # Hardcoded to use llama3-8b-8192
-            if not llm:
-                # Error is already shown by get_llm, but update status_placeholder too
-                status_placeholder.error("Failed to initialize LLM. Application cannot proceed.")
-                st.stop()
-
-            end_time = time.time()
-            # status_placeholder is updated by get_llm or on success below
-            status_placeholder.success(f"Application Ready! (Loaded in {end_time - start_time:.2f}s)")
-
-        retriever = vector_store.as_retriever(search_kwargs={"k": 5})
-
-        # --- Query Input and Response ---
-        st.markdown("---")
-        st.subheader("Ask a question about our documents:")
-
-
-
-        if "messages" not in st.session_state:
-            st.session_state.messages = []
-
-        query = st.text_input("Enter your question:", key="query_input", placeholder="e.g., 'What is the return policy?' or 'Status of order for John Doe?'")
-
-        if st.button("Submit", key="submit_button"):
-            if query:
-                st.session_state.messages.append({"role": "user", "content": query})
-
-                current_model_info = st.sidebar.empty()  # Placeholder for current mode info
-
-                if "order" in query.lower() and (
-                        "status" in query.lower() or "track" in query.lower() or "update" in query.lower() or any(
-                        name_part.lower() in query.lower() for name_part in ["customer", "client", "name"])):
-                    active_prompt_template = ORDER_STATUS_PROMPT
-                    current_model_info.info("Mode: Order Status Query")
-                else:
-                    active_prompt_template = GENERAL_QA_PROMPT
-                    current_model_info.info("Mode: General Query")
-
-                rag_chain = get_rag_chain(llm, retriever, active_prompt_template)
-
-                with st.spinner("Thinking..."):
-                    try:
-                        response = rag_chain.invoke(query)
-                        st.session_state.messages.append({"role": "assistant", "content": response})
-                    except Exception as e:
-                        st.error(f"Error during RAG chain invocation: {e}")
-                        response = "Sorry, I encountered an error while processing your request."
-                        st.session_state.messages.append({"role": "assistant", "content": response})
-            else:
-                st.warning("Please enter a question.")
+    # --- Knowledge Base Loading ---
+    with st.spinner("Knowledge Base is loading... Please wait."):
+        start_time = time.time()
+        processed_documents = load_and_process_documents(DOCS_DIR)
+        if not processed_documents:
+            status_placeholder.error("Failed to load or process documents. Check logs and `docs` folder.")
+            st.stop()
+
+        vector_store = create_vector_store(processed_documents, EMBEDDING_MODEL_NAME)
+        if not vector_store:
+            status_placeholder.error("Failed to create vector store. Application cannot proceed.")
+            st.stop()
+
+        # Pass the selected model to get_llm
+        llm = get_llm(groq_api_key, model_name="llama3-8b-8192") # Hardcoded to use llama3-8b-8192
+        if not llm:
+            # Error is already shown by get_llm, but update status_placeholder too
+            status_placeholder.error("Failed to initialize LLM. Application cannot proceed.")
+            st.stop()
+
+        end_time = time.time()
+        # status_placeholder is updated by get_llm or on success below
+        status_placeholder.success(f"Application Ready! (Loaded in {end_time - start_time:.2f}s)")
+
+    retriever = vector_store.as_retriever(search_kwargs={"k": 5})
 
-        st.markdown("---")
-        st.subheader("Response:")
-        response_area = st.container()
-        # Ensure response_area is robust against empty messages or incorrect last role
-        last_assistant_message = "Ask a question to see the answer here."
-        if st.session_state.messages and st.session_state.messages[-1]['role'] == 'assistant':
-            last_assistant_message = st.session_state.messages[-1]['content']
+    # --- Query Input and Response ---
+
+    st.markdown("---")
+    st.subheader("Ask a question about our documents:")
+
+    # Prompt templates
+    GENERAL_QA_PROMPT = """
+    You are an AI assistant for our internal knowledge base.
+    Your goal is to provide accurate and concise answers based ONLY on the provided context.
+    Do not make up information. If the answer is not found in the context, state that clearly.
+    Ensure your answers are directly supported by the text.
+    Accuracy is paramount.
+
+    Context:
+    {context}
+
+    Question: {question}
+
+    Answer:
+    """
+
+    ORDER_STATUS_PROMPT = """
+    You are an AI assistant helping with customer order inquiries.
+    Based ONLY on the following retrieved information from our order system and policies:
+    {context}
+
+    The customer's query is: {question}
+
+    Please perform the following steps:
+    1. Carefully analyze the context for any order details (Order ID, Customer Name, Status, Items, Dates, etc.).
+    2. If an order matching the query (or related to a name in the query) is found in the context:
+        - Address the customer by their name if available in the order details (e.g., "Hello [Customer Name],").
+        - Provide ALL available information about their order, including Order ID, status, items, dates, and any other relevant details found in the context.
+        - Be comprehensive and clear.
+    3. If no specific order details are found in the context that match the query, or if the context is insufficient, politely state that you couldn't find the specific order information in the provided documents and suggest they contact support for further assistance.
+    4. Do NOT invent or infer any information not explicitly present in the context.
+
+    Answer:
+    """
+
+    if "messages" not in st.session_state:
+        st.session_state.messages = []
+
+    query = st.text_input("Enter your question:", key="query_input", placeholder="e.g., 'What is the return policy?' or 'Status of order for John Doe?'")
+
+    if st.button("Submit", key="submit_button"):
+        if query:
+            st.session_state.messages.append({"role": "user", "content": query})
+
+            current_model_info = st.sidebar.empty() # Placeholder for current mode info
+
+            if "order" in query.lower() and ("status" in query.lower() or "track" in query.lower() or "update" in query.lower() or any(name_part.lower() in query.lower() for name_part in ["customer", "client", "name"])):
+                active_prompt_template = ORDER_STATUS_PROMPT
+                current_model_info.info("Mode: Order Status Query")
+            else:
+                active_prompt_template = GENERAL_QA_PROMPT
+                current_model_info.info("Mode: General Query")
+
+            rag_chain = get_rag_chain(llm, retriever, active_prompt_template)
+
+            with st.spinner("Thinking..."):
+                try:
+                    response = rag_chain.invoke(query)
+                    st.session_state.messages.append({"role": "assistant", "content": response})
+                except Exception as e:
+                    st.error(f"Error during RAG chain invocation: {e}")
+                    response = "Sorry, I encountered an error while processing your request."
+                    st.session_state.messages.append({"role": "assistant", "content": response})
+        else:
+            st.warning("Please enter a question.")
+
+    st.markdown("---")
+    st.subheader("Response:")
+    response_area = st.container()
+    # Ensure response_area is robust against empty messages or incorrect last role
+    last_assistant_message = "Ask a question to see the answer here."
+    if st.session_state.messages and st.session_state.messages[-1]['role'] == 'assistant':
+        last_assistant_message = st.session_state.messages[-1]['content']
+
+    response_area.markdown(f"<div class='response-area'>{last_assistant_message}</div>", unsafe_allow_html=True)
 
-        response_area.markdown(f"<div class='response-area'>{last_assistant_message}</div>", unsafe_allow_html=True)
 
     st.sidebar.markdown("---")
     st.sidebar.markdown("Built with ❤️ using Streamlit & Langchain & Groq")