Rulga/LS-chatbot-log

1import os2import time3import streamlit as st4from dotenv import load_dotenv5from langchain_groq import ChatGroq6from langchain_huggingface import HuggingFaceEmbeddings7from langchain_community.vectorstores import FAISS8from langchain_text_splitters import RecursiveCharacterTextSplitter9from langchain_community.document_loaders import WebBaseLoader10from langchain_core.prompts import PromptTemplate11from langchain_core.output_parsers import StrOutputParser12from langchain_core.runnables import RunnableLambda13import requests14import json15 16# Логирует взаимодействие в JSON-файл17from datetime import datetime18 19 20def log_interaction(user_input: str, bot_response: str):21    """Логирует взаимодействие в JSON-файл"""22    log_entry = {23        "timestamp": datetime.now().isoformat(),24        "user_input": user_input,25        "bot_response": bot_response26    }27    28    log_dir = "chat_history"29    os.makedirs(log_dir, exist_ok=True)30    31    log_path = os.path.join(log_dir, "chat_logs.json")32    with open(log_path, "a") as f:33        f.write(json.dumps(log_entry) + "\n")34 35#36 37 38 39# Page configuration40st.set_page_config(page_title="Status Law Assistant", page_icon="⚖️")41 42# Knowledge base info in session_state43if 'kb_info' not in st.session_state:44    st.session_state.kb_info = {45        'build_time': None,46        'size': None47    }48 49# Display title and knowledge base info50# st.title("www.Status.Law Legal Assistant")51 52st.markdown(53    '''54    <h1>55        ⚖️ 56        <a href="https://status.law/" style="text-decoration: underline; color: blue; font-size: inherit;">57            Status.Law58        </a> 59        Legal Assistant60    </h1>61    ''',62    unsafe_allow_html=True63)64 65if st.session_state.kb_info['build_time'] and st.session_state.kb_info['size']:66    st.caption(f"(Knowledge base build time: {st.session_state.kb_info['build_time']:.2f} seconds, "67               f"size: {st.session_state.kb_info['size']:.2f} MB)")68 69# Path to store vector database70VECTOR_STORE_PATH = "vector_store"71 72# Создание папки истории, если она не существует73if not os.path.exists("chat_history"):74    os.makedirs("chat_history")75 76# Website URLs77urls = [78    "https://status.law",  79    "https://status.law/about",80    "https://status.law/careers",  81    "https://status.law/tariffs-for-services-of-protection-against-extradition",82    "https://status.law/challenging-sanctions",83    "https://status.law/law-firm-contact-legal-protection"84    "https://status.law/cross-border-banking-legal-issues", 85    "https://status.law/extradition-defense", 86    "https://status.law/international-prosecution-protection", 87    "https://status.law/interpol-red-notice-removal",  88    "https://status.law/practice-areas",  89    "https://status.law/reputation-protection",90    "https://status.law/faq"91]92 93# Load secrets94try:95    GROQ_API_KEY = st.secrets["GROQ_API_KEY"]96except Exception as e:97    st.error("Error loading secrets. Please check your configuration.")98    st.stop()99 100# Initialize models101@st.cache_resource102def init_models():103    llm = ChatGroq(104        model_name="llama-3.3-70b-versatile",105        temperature=0.6,106        api_key=GROQ_API_KEY107    )108    embeddings = HuggingFaceEmbeddings(109        model_name="intfloat/multilingual-e5-large-instruct"110    )111    return llm, embeddings112 113# Build knowledge base114def build_knowledge_base(embeddings):115    start_time = time.time()116    117    documents = []118    with st.status("Loading website content...") as status:119        for url in urls:120            try:121                loader = WebBaseLoader(url)122                docs = loader.load()123                documents.extend(docs)124                status.update(label=f"Loaded {url}")125            except Exception as e:126                st.error(f"Error loading {url}: {str(e)}")127                128    text_splitter = RecursiveCharacterTextSplitter(129        chunk_size=500,130        chunk_overlap=100131    )132    chunks = text_splitter.split_documents(documents)133    134    vector_store = FAISS.from_documents(chunks, embeddings)135    vector_store.save_local(VECTOR_STORE_PATH)136    137    end_time = time.time()138    build_time = end_time - start_time139    140    # Calculate knowledge base size141    total_size = 0142    for path, dirs, files in os.walk(VECTOR_STORE_PATH):143        for f in files:144            fp = os.path.join(path, f)145            total_size += os.path.getsize(fp)146    size_mb = total_size / (1024 * 1024)147    148    # Save knowledge base info149    st.session_state.kb_info['build_time'] = build_time150    st.session_state.kb_info['size'] = size_mb151    152    st.success(f"""153    Knowledge base created successfully:154    - Time taken: {build_time:.2f} seconds155    - Size: {size_mb:.2f} MB156    - Number of chunks: {len(chunks)}157    """)158    159    return vector_store160 161# Main function162def main():163    # Initialize models164    llm, embeddings = init_models()165    166    # Check if knowledge base exists167    if not os.path.exists(VECTOR_STORE_PATH):168        st.warning("Knowledge base not found.")169        if st.button("Create Knowledge Base"):170            vector_store = build_knowledge_base(embeddings)171            st.session_state.vector_store = vector_store172            st.rerun()173    else:174        if 'vector_store' not in st.session_state:175            st.session_state.vector_store = FAISS.load_local(176                VECTOR_STORE_PATH,177                embeddings,178                allow_dangerous_deserialization=True179            )180    181    # Chat mode182    if 'vector_store' in st.session_state:183        if 'messages' not in st.session_state:184            st.session_state.messages = []185            186        # Display chat history187        for message in st.session_state.messages:188            st.chat_message("user").write(message["question"])189            st.chat_message("assistant").write(message["answer"])190            191        # User input192        if question := st.chat_input("Ask your question"):193            st.chat_message("user").write(question)194            195            # Retrieve context and generate response196            with st.chat_message("assistant"):197                with st.spinner("Thinking..."):198                    context = st.session_state.vector_store.similarity_search(question)199                    context_text = "\n".join([doc.page_content for doc in context])200                    201                    prompt = PromptTemplate.from_template("""202                    You are a helpful and polite legal assistant at Status Law.203                    You answer in the language in which the question was asked.204                    Answer the question based on the context provided.205                    If you cannot answer based on the context, say so politely and offer to contact Status Law directly via the following channels:206                    - For all users: +32465594521 (landline phone).207                    - For English and Swedish speakers only: +46728495129 (available on WhatsApp, Telegram, Signal, IMO).208                    - Provide a link to the contact form: [Contact Form](https://status.law/law-firm-contact-legal-protection/).209                    If the user has questions about specific services and their costs, suggest they visit the page https://status.law/tariffs-for-services-of-protection-against-extradition-and-international-prosecution/ for detailed information.210 211                    Ask the user additional questions to understand which service to recommend and provide an estimated cost. For example, clarify their situation and needs to suggest the most appropriate options.212 213                    Also, offer free consultations if they are available and suitable for the user's request.214                    Answer professionally but in a friendly manner.215 216                    Example:217                    Q: How can I challenge the sanctions?218                    A: To challenge the sanctions, you should consult with our legal team, who specialize in this area. Please contact us directly for detailed advice. You can fill out our contact form here: [Contact Form](https://status.law/law-firm-contact-legal-protection/).219 220                    Context: {context}221                    Question: {question}222                    """)223                    224                    chain = prompt | llm | StrOutputParser()225                    response = chain.invoke({226                        "context": context_text,227                        "question": question228                    })229                    230                    st.write(response)231                    232                    233                    # В блоке генерации ответа (после st.write(response))   234                    log_interaction(question, response)235                    # Save chat history236                    st.session_state.messages.append({237                        "question": question,238                        "answer": response239                    })240 241if __name__ == "__main__":242    main()243