import streamlit as st from langchain_community.document_loaders import PyPDFLoader from langchain_text_splitters import RecursiveCharacterTextSplitter from langchain_ollama import OllamaEmbeddings from langchain_ollama import ChatOllama from langchain_community.vectorstores import Chroma from langchain.prompts import ChatPromptTemplate import os import tempfile # Set page configuration st.set_page_config( page_title="PDF Chatbot", page_icon="📚", layout="wide" ) # App title and description st.title("📚 Multi-PDF Chatbot") st.markdown("Upload multiple PDF documents and ask questions about their content.") # Initialize session state variables if "conversation_history" not in st.session_state: st.session_state.conversation_history = [] if "vectorstore" not in st.session_state: st.session_state.vectorstore = None if "uploaded_files" not in st.session_state: st.session_state.uploaded_files = [] if "processing_complete" not in st.session_state: st.session_state.processing_complete = False # Sidebar for file upload and processing with st.sidebar: st.header("Document Upload") uploaded_files = st.file_uploader("Upload PDF Documents", type="pdf", accept_multiple_files=True) # Process button if uploaded_files and st.button("Process Documents"): with st.spinner("Processing PDFs..."): # Reset processing state st.session_state.processing_complete = False st.session_state.conversation_history = [] # Create a temporary directory to save uploaded files temp_dir = tempfile.mkdtemp() all_docs = [] # Save uploaded files to temp directory and process them for file in uploaded_files: temp_file_path = os.path.join(temp_dir, file.name) with open(temp_file_path, "wb") as f: f.write(file.getvalue()) st.write(f"Loading {file.name}...") loader = PyPDFLoader(temp_file_path) docs = loader.load() all_docs.extend(docs) st.session_state.uploaded_files = [file.name for file in uploaded_files] st.write(f"Loaded {len(all_docs)} pages in total.") # Split documents into chunks text_splitter = RecursiveCharacterTextSplitter( chunk_size=1000, chunk_overlap=200 ) chunks = text_splitter.split_documents(all_docs) st.write(f"Split into {len(chunks)} chunks") # Create embeddings and store in vector database with st.spinner("Creating embeddings... this might take a while"): try: embeddings = OllamaEmbeddings(model="nomic-embed-text") st.session_state.vectorstore = Chroma.from_documents(chunks, embeddings) st.session_state.processing_complete = True st.success("Documents processed successfully!") except Exception as e: st.error(f"Error creating embeddings: {str(e)}") # Display processed files if st.session_state.uploaded_files: st.header("Processed Documents") for file_name in st.session_state.uploaded_files: st.write(f"- {file_name}") # Model settings st.header("Model Settings") model_name = st.selectbox( "Select Ollama Model", ["llama3:8b", "deepseek-r1:8b"], index=0 ) # Main chat interface st.header("Chat with your PDFs") # Display conversation history for i, (query, response) in enumerate(st.session_state.conversation_history): # User message with st.chat_message("user"): st.write(query) # Bot message with st.chat_message("assistant"): st.write(response) # Chat input user_query = st.chat_input("Ask a question about your PDFs...", disabled=not st.session_state.processing_complete) if user_query and st.session_state.processing_complete: # Display user message with st.chat_message("user"): st.write(user_query) # Generate response with st.chat_message("assistant"): with st.spinner("Thinking..."): # Define helper functions for generating responses def format_docs(docs): return "\n\n".join(doc.page_content for doc in docs) def format_chat_history(history): if not history: return "No previous conversation." formatted = "" for i, (q, a) in enumerate(history): formatted += f"Question {i+1}: {q}\nAnswer {i+1}: {a}\n\n" return formatted # Create LLM instance llm = ChatOllama( base_url='http://127.0.0.1:11434', model=model_name ) # Set up retriever retriever = st.session_state.vectorstore.as_retriever(search_kwargs={"k": 4}) # Create prompt template template = """ You are a helpful assistant that answers questions based on the provided PDF documents. Answer the question based only on the following context from the PDFs: {context} Previous conversation history: {chat_history} Current question: {question} Provide a comprehensive and accurate answer based on the information in the PDFs. """ prompt = ChatPromptTemplate.from_template(template) try: # Retrieve relevant documents docs = retriever.get_relevant_documents(user_query) context = format_docs(docs) # Format conversation history chat_history = format_chat_history(st.session_state.conversation_history) # Generate response messages = prompt.format_messages( context=context, chat_history=chat_history, question=user_query ) response = llm.invoke(messages).content # Display response st.write(response) # Update conversation history st.session_state.conversation_history.append((user_query, response)) except Exception as e: st.error(f"Error generating response: {str(e)}") # Show a message if no documents have been processed if not st.session_state.processing_complete: st.info("👈 Please upload and process PDF documents from the sidebar to start chatting.") # Add a clear chat button if st.session_state.conversation_history and st.button("Clear Chat History"): st.session_state.conversation_history = [] st.experimental_rerun() # Add footer st.markdown("---") st.markdown("Built with Langchain and Ollama")