app.py

# 第二步：匯入必要的庫並初始化
import os
import gradio as gr
from PyPDF2 import PdfReader
from langchain.text_splitter import CharacterTextSplitter
from langchain_google_genai import GoogleGenerativeAIEmbeddings, ChatGoogleGenerativeAI
from langchain_community.vectorstores import FAISS
from langchain.chains.question_answering import load_qa_chain
from langchain.prompts import PromptTemplate
import shutil
import tempfile
from docx import Document
from docx.shared import Inches
from datetime import datetime
import resend

print("📦 所有庫匯入成功！")

# 第三步：設置API密鑰
print("🔑 設置API密鑰...")

# Gemini API key (請替換為您自己的API密鑰)
gemini_api_key = "AIzaSyBbufVdrxdZkBxXLzXxfdtGArHUMfos5Z0"
os.environ["GOOGLE_API_KEY"] = gemini_api_key

# Resend API key (請替換為您自己的API密鑰)
resend.api_key = "re_TPd7f23i_E3gvJYJF8xibuymWSPXxKPrY"

print("✅ API密鑰設置完成！")

# 第四步：定義PDF聊天機器人類
class PDFChatBot:
    def __init__(self):
        self.vector_store = None
        self.embeddings = GoogleGenerativeAIEmbeddings(
            model="models/text-embedding-004",
            google_api_key=gemini_api_key
        )
        self.processed_files = []
        self.chat_history = []

    def get_pdf_text(self, pdf_files):
        """從多個PDF檔案中提取文字"""
        raw_text = ""
        processed_count = 0

        if not pdf_files:
            return raw_text, processed_count

        if not isinstance(pdf_files, list):
            pdf_files = [pdf_files]

        for pdf_file in pdf_files:
            try:
                pdf_path = pdf_file.name if hasattr(pdf_file, 'name') else pdf_file
                pdf_reader = PdfReader(pdf_path)
                file_text = ""
                for page in pdf_reader.pages:
                    text = page.extract_text()
                    if text:
                        file_text += text + "\n"

                if file_text.strip():
                    raw_text += file_text
                    processed_count += 1
                    self.processed_files.append(os.path.basename(pdf_path))

            except Exception as e:
                print(f"讀取PDF時發生錯誤：{str(e)}")
                continue

        return raw_text, processed_count

    def get_text_chunks(self, text):
        """將文字分割成區塊進行處理"""
        text_splitter = CharacterTextSplitter(
            separator="\n",
            chunk_size=10000,
            chunk_overlap=1000,
            length_function=len
        )
        chunks = text_splitter.split_text(text)
        return chunks

    def create_vector_store(self, chunks):
        """從文字區塊建立FAISS向量儲存"""
        try:
            self.vector_store = FAISS.from_texts(chunks, self.embeddings)
            self.vector_store.save_local("faiss_index")
            return True
        except Exception as e:
            print(f"建立向量儲存時發生錯誤：{str(e)}")
            return False

    def load_vector_store(self):
        """載入已存在的向量儲存"""
        try:
            if os.path.exists("faiss_index"):
                self.vector_store = FAISS.load_local(
                    "faiss_index",
                    embeddings=self.embeddings,
                    allow_dangerous_deserialization=True
                )
                return True
            else:
                return False
        except Exception as e:
            print(f"載入向量儲存時發生錯誤：{str(e)}")
            return False

    def get_conversational_chain(self):
        """建立對話鏈"""
        prompt_template = """
        根據提供的內容盡可能詳細地回答問題。確保提供所有細節。
        如果你需要更多細節來完美回答問題，那麼請詢問你認為需要了解的更多細節。
        如果答案不在提供的內容中，只需說"在您提供的內容中找不到答案"。不要提供錯誤的答案。

        內容:\n {context}\n
        問題: \n{question}\n

        回答:
        """

        model = ChatGoogleGenerativeAI(
            model="gemini-2.0-flash-exp",
            google_api_key=gemini_api_key,
            temperature=0.3,
            max_tokens=8192,
            top_p=0.8,
            top_k=40
        )

        prompt = PromptTemplate(
            template=prompt_template,
            input_variables=['context', 'question']
        )

        chain = load_qa_chain(model, chain_type="stuff", prompt=prompt)
        return chain

    def answer_question(self, question):
        """回答使用者問題"""
        if not self.vector_store:
            return "請先上傳並處理PDF檔案！"

        if not question.strip():
            return "請輸入您的問題。"

        try:
            docs = self.vector_store.similarity_search(question, k=6)
            if not docs:
                return "在上傳的文件中找不到相關資訊。"

            chain = self.get_conversational_chain()
            response = chain(
                {
                    "input_documents": docs,
                    "question": question,
                },
                return_only_outputs=True
            )

            return response["output_text"]

        except Exception as e:
            return f"處理問題時發生錯誤：{str(e)}"

    def process_pdfs(self, pdf_files, progress=gr.Progress()):
        """處理PDF檔案"""
        if not pdf_files:
            return "請上傳至少一個PDF檔案。", ""

        self.processed_files = []
        progress(0, desc="開始處理PDF檔案...")

        progress(0.2, desc="提取PDF文字內容...")
        raw_text, processed_count = self.get_pdf_text(pdf_files)

        if not raw_text.strip():
            return "無法從PDF檔案中提取到文字。", ""

        progress(0.4, desc="分割文字內容...")
        text_chunks = self.get_text_chunks(raw_text)

        progress(0.6, desc="建立向量儲存...")
        success = self.create_vector_store(text_chunks)

        progress(1.0, desc="處理完成!")

        if success:
            file_list = "已處理的檔案:\n" + "\n".join([f"• {file}" for file in self.processed_files])
            return f"✅ 成功處理 {processed_count} 個PDF檔案！\n總共 {len(text_chunks)} 個文字區塊\n現在您可以開始提問。", file_list
        else:
            return "❌ PDF處理失敗，請重試。", ""

    def clear_data(self):
        """清除處理過的資料"""
        try:
            if os.path.exists("faiss_index"):
                shutil.rmtree("faiss_index")
            self.vector_store = None
            self.processed_files = []
            self.chat_history = []
            return "✅ 已清除所有處理過的資料！", ""
        except Exception as e:
            return f"❌ 清除資料時發生錯誤：{str(e)}", ""

    def create_docx_report(self, chat_history):
        """建立包含聊天記錄的docx報告"""
        try:
            doc = Document()
            title = doc.add_heading('PDF聊天機器人 - 問答記錄', 0)
            title.alignment = 1

            doc.add_paragraph(f'產生時間：{datetime.now().strftime("%Y年%m月%d日 %H:%M:%S")}')

            if self.processed_files:
                doc.add_heading('已處理的PDF檔案：', level=2)
                for i, file in enumerate(self.processed_files, 1):
                    doc.add_paragraph(f'{i}. {file}', style='List Number')

            doc.add_paragraph('')
            doc.add_heading('問答記錄：', level=2)

            if not chat_history:
                doc.add_paragraph('目前沒有問答記錄。')
            else:
                for i in range(0, len(chat_history), 2):
                    if i + 1 < len(chat_history):
                        question = chat_history[i]['content']
                        answer = chat_history[i + 1]['content']

                        q_paragraph = doc.add_paragraph()
                        q_run = q_paragraph.add_run(f'問題 {(i//2)+1}：')
                        q_run.bold = True
                        q_run.font.size = Inches(0.14)
                        q_paragraph.add_run(question)

                        a_paragraph = doc.add_paragraph()
                        a_run = a_paragraph.add_run('回答：')
                        a_run.bold = True
                        a_run.font.size = Inches(0.14)
                        a_paragraph.add_run(answer)

                        doc.add_paragraph('─' * 50)

            temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.docx')
            doc.save(temp_file.name)
            temp_file.close()

            return temp_file.name

        except Exception as e:
            print(f"建立docx檔案時發生錯誤：{str(e)}")
            return None

    def create_email_html_content(self, chat_history):
        """建立郵件的HTML內容"""
        if not chat_history:
            return "<p>目前沒有問答記錄。</p>"

        html_content = f"""
        <html>
        <head>
            <style>
                body {{ font-family: Arial, sans-serif; line-height: 1.6; color: #333; }}
                .header {{ background-color: #f4f4f4; padding: 20px; text-align: center; }}
                .content {{ padding: 20px; }}
                .question {{ background-color: #e8f4f8; padding: 10px; margin: 10px 0; border-left: 4px solid #2196F3; }}
                .answer {{ background-color: #f0f8e8; padding: 10px; margin: 10px 0; border-left: 4px solid #4CAF50; }}
                .file-list {{ background-color: #fff3cd; padding: 10px; margin: 10px 0; border: 1px solid #ffeeba; }}
                hr {{ border: none; border-top: 1px solid #ddd; margin: 20px 0; }}
            </style>
        </head>
        <body>
            <div class="header">
                <h1>🤖 PDF聊天機器人 - 問答記錄</h1>
                <p>產生時間：{datetime.now().strftime("%Y年%m月%d日 %H:%M:%S")}</p>
            </div>
            <div class="content">
        """

        if self.processed_files:
            html_content += """
                <div class="file-list">
                    <h3>📁 已處理的PDF檔案：</h3>
                    <ul>
            """
            for file in self.processed_files:
                html_content += f"<li>{file}</li>"
            html_content += "</ul></div>"

        html_content += "<h3>💬 問答記錄：</h3>"

        for i in range(0, len(chat_history), 2):
            if i + 1 < len(chat_history):
                question = chat_history[i]['content']
                answer = chat_history[i + 1]['content']

                question_html = question.replace('\n', '<br>')
                answer_html = answer.replace('\n', '<br>')

                html_content += f"""
                    <div class="question">
                        <strong>問題 {(i//2)+1}：</strong><br>
                        {question_html}
                    </div>
                    <div class="answer">
                        <strong>回答：</strong><br>
                        {answer_html}
                    </div>
                    <hr>
                """

        html_content += """
            </div>
        </body>
        </html>
        """

        return html_content

    def send_chat_history_email(self, recipient_email):
        """發送聊天記錄到指定信箱"""
        if not self.chat_history:
            return "❌ 沒有聊天記錄可以發送！"

        if not recipient_email or "@" not in recipient_email:
            return "❌ 請輸入有效的信箱地址！"

        try:
            html_content = self.create_email_html_content(self.chat_history)

            r = resend.Emails.send({
                "from": "onboarding@resend.dev",
                "to": recipient_email,
                "subject": f"PDF聊天機器人問答記錄 - {datetime.now().strftime('%Y-%m-%d %H:%M')}",
                "html": html_content
            })

            return f"✅ 郵件已成功發送到 {recipient_email}！\n郵件ID: {r.get('id', 'Unknown')}"

        except Exception as e:
            return f"❌ 發送郵件時發生錯誤：{str(e)}"

# 第五步：初始化聊天機器人
print("🤖 初始化PDF聊天機器人...")
bot = PDFChatBot()

# 第六步：定義Gradio介面函數
def upload_and_process(files, progress=gr.Progress()):
    return bot.process_pdfs(files, progress)

def ask_question(question, history):
    if not question.strip():
        return history, ""

    response = bot.answer_question(question)
    user_msg = {"role": "user", "content": question}
    assistant_msg = {"role": "assistant", "content": response}

    history.append(user_msg)
    history.append(assistant_msg)
    bot.chat_history = history.copy()

    return history, ""

def download_chat_history():
    if not bot.chat_history:
        return None
    docx_path = bot.create_docx_report(bot.chat_history)
    return docx_path

def send_email(email_address):
    return bot.send_chat_history_email(email_address)

def clear_chat():
    bot.chat_history = []
    return [], ""

def clear_all_data():
    return bot.clear_data()

def load_existing_data():
    if bot.load_vector_store():
        return "✅ 成功載入已處理的資料！", ""
    else:
        return "❌ 沒有找到已處理的資料。", ""

# 第七步：建立Gradio介面
print("🎨 建立使用者介面...")

with gr.Blocks(title="PDF聊天機器人", theme=gr.themes.Soft()) as demo:
    gr.Markdown(
        """
        # 🤖 PDF聊天機器人 (Flash 2.0 + 郵件發送)

        上傳您的PDF檔案，然後就可以向文件提問！支援多語言問答並可將記錄發送到信箱。

        **🔥 在Hugging Face中執行**
        """
    )

    with gr.Tab("📁 檔案處理"):
        with gr.Row():
            with gr.Column(scale=2):
                file_upload = gr.File(
                    file_count="multiple",
                    file_types=[".pdf"],
                    label="上傳PDF檔案",
                    height=200
                )

                with gr.Row():
                    process_btn = gr.Button("🚀 處理PDF檔案", variant="primary", size="lg")
                    load_btn = gr.Button("📂 載入已處理資料", variant="secondary")
                    clear_btn = gr.Button("🗑️ 清除資料", variant="stop")

            with gr.Column(scale=1):
                status_text = gr.Textbox(
                    label="處理狀態",
                    lines=8,
                    interactive=False
                )

                file_list = gr.Textbox(
                    label="已處理檔案",
                    lines=6,
                    interactive=False
                )

    with gr.Tab("💬 問答聊天"):
        chatbot = gr.Chatbot(
            label="聊天記錄",
            height=500,
            show_copy_button=True,
            type="messages"
        )

        with gr.Row():
            question_input = gr.Textbox(
                placeholder="請輸入您的問題...",
                label="問題",
                lines=2,
                scale=4
            )
            ask_btn = gr.Button("📤 提問", variant="primary", scale=1)

        with gr.Row():
            clear_chat_btn = gr.Button("🧹 清除聊天記錄", variant="secondary", scale=1)
            download_btn = gr.Button("📥 下載問答記錄", variant="primary", scale=1)

        download_file = gr.File(visible=False)

        gr.Examples(
            examples=[
                "這份文件的主要內容是什麼？",
                "請總結文件的重點。",
                "文件中提到了哪些重要概念？",
                "能否詳細解釋某個特定主題？"
            ],
            inputs=question_input,
            label="問題範例"
        )

    with gr.Tab("📧 郵件發送"):
        gr.Markdown(
            """
            ### 📮 發送聊天記錄到信箱

            將您的問答記錄以精美的HTML格式發送到指定信箱，方便保存和分享。
            """
        )

        with gr.Row():
            with gr.Column(scale=2):
                email_input = gr.Textbox(
                    label="收件人信箱",
                    placeholder="請輸入有效的信箱地址...",
                    value="grace.chenyiwen@gmail.com"
                )

                send_email_btn = gr.Button("📧 發送聊天記錄", variant="primary", size="lg")

            with gr.Column(scale=1):
                email_status = gr.Textbox(
                    label="發送狀態",
                    lines=6,
                    interactive=False
                )

        gr.Markdown(
            """
            **注意事項：**
            - 請確保您已經有一些問答記錄
            - 郵件將包含所有處理過的PDF檔案清單和完整的問答記錄
            - 郵件格式為HTML，在大多數郵件用戶端中都能正常顯示
            """
        )

    # 事件處理
    def handle_download():
        file_path = download_chat_history()
        if file_path:
            return gr.update(value=file_path, visible=True)
        else:
            gr.Warning("沒有聊天記錄可以下載！")
            return gr.update(visible=False)

    # 綁定事件
    process_btn.click(
        fn=upload_and_process,
        inputs=[file_upload],
        outputs=[status_text, file_list],
        show_progress=True
    )

    load_btn.click(
        fn=load_existing_data,
        outputs=[status_text, file_list]
    )

    clear_btn.click(
        fn=clear_all_data,
        outputs=[status_text, file_list]
    )

    ask_btn.click(
        fn=ask_question,
        inputs=[question_input, chatbot],
        outputs=[chatbot, question_input]
    )

    question_input.submit(
        fn=ask_question,
        inputs=[question_input, chatbot],
        outputs=[chatbot, question_input]
    )

    clear_chat_btn.click(
        fn=clear_chat,
        outputs=[chatbot, question_input]
    )

    download_btn.click(
        fn=handle_download,
        outputs=download_file
    )

    send_email_btn.click(
        fn=send_email,
        inputs=[email_input],
        outputs=[email_status]
    )

# 第八步：啟動應用程式
print("🚀 啟動應用程式中...")

# 嘗試載入現有的向量儲存
bot.load_vector_store()

# 在Hugging Face中啟動應用程式
demo.launch(
    share=True,  # 在Hugging Face中設為True獲得公共連結
    server_name="0.0.0.0",  # 允許外部存取
    server_port=None,
    show_error=True,
    debug=True
)

print("✅ PDF聊天機器人已成功啟動！")
print("📍 請點選上方顯示的連結來存取應用程式")