update

README.md

@@ -0,0 +1,15 @@
+---
+title: FAQ Workshop
+emoji: 💻
+colorFrom: gray
+colorTo: purple
+sdk: gradio
+sdk_version: 5.16.0
+python_version: 3.11.11
+app_file: app.py
+pinned: false
+license: mit
+short_description: FAQ CHAT by プロジェクトマネジメント勉強会
+---
+
+Check out the configuration reference at <https://huggingface.co/docs/hub/spaces-config-reference>

app.py

@@ -0,0 +1,299 @@
+import base64
+import os
+import shutil
+import threading
+import time
+
+import gradio as gr
+
+from blog_class import audio_text, knowledge_class, save_feedback
+
+knowledge_cls = knowledge_class()
+
+
+def update_knowledge():
+    knowledge_cls.get_new_knowledge()
+    choices_list = list(knowledge_cls.reference_dict.keys())
+
+    return gr.update(choices=choices_list)
+
+
+def user_message_fn(user_message, history):
+    if history is None:
+        history = []
+    history.append({"role": "user", "content": user_message})
+    return "", history
+
+
+def answer_question(history):
+    print("chat Start")
+    answer_text, info_title, audio_file_path = knowledge_cls.get_chat_answer(history)
+    if not answer_text:
+        yield history
+        return
+    audio_answer_container = {"result": None}
+
+    def run_audio_answer():
+        audio_answer_container["result"] = audio_text(answer_text, audio_file_path)
+
+    audio_thread = threading.Thread(target=run_audio_answer)
+    audio_thread.start()
+
+    history.append({"role": "assistant", "content": ""})
+    #
+    history.append(
+        {
+            "role": "assistant",
+            "content": "音声作成中...",
+            "metadata": {"title": "🎧 音声返信", "status": "pending"},
+        }
+    )
+
+    history.append(
+        {
+            "role": "assistant",
+            "content": info_title,
+            "metadata": {"title": "📚 参照ナレッジ", "status": "done"},
+        }
+    )
+
+    for char in answer_text:
+        history[-3]["content"] += char
+        if (
+            audio_answer_container["result"] is not None
+            and history[-2]["metadata"]["status"] != "done"
+        ):
+
+            audio_str = base64.b64encode(
+                open(audio_answer_container["result"], "rb").read()
+            ).decode("utf-8")
+
+            history[-2]["content"] = gr.HTML(
+                f"""
+                <audio controls>
+                    <source src="data:audio/mp3;base64,{audio_str}" type="audio/mp3">
+                </audio>
+                """
+            )
+
+            history[-2]["metadata"] = {"title": "🎧 音声返信", "status": "done"}
+
+        yield history
+        time.sleep(0.03)
+    while audio_thread.is_alive():
+        yield history
+        time.sleep(0.2)
+    audio_thread.join()
+    print("chat end")
+    if history[-2]["metadata"]["status"] != "done":
+        audio_str = base64.b64encode(open(audio_file_path, "rb").read()).decode("utf-8")
+        history[-2]["content"] = gr.HTML(
+            f"""
+            <audio controls>
+                <source src="data:audio/mp3;base64,{audio_str}" type="audio/mp3">
+            </audio>
+            """
+        )
+        history[-2]["metadata"] = {"title": "🎧 音声返信", "status": "done"}
+    yield history
+
+
+def handle_feedback(like_data: gr.LikeData):
+    save_feedback(like_data.value[0], like_data.liked)
+
+
+def get_reference_info(selected_reference):
+    reference_dict = knowledge_cls.reference_dict.get(selected_reference, None)
+    if reference_dict:
+        yield (
+            reference_dict["audio"],
+            reference_dict["original_text"],
+            reference_dict["summary"],
+        )
+    else:
+        yield (None, None, None)
+
+
+def zip_directory():
+    zip_path = "resource.zip"
+    if os.path.exists(zip_path):
+        os.remove(zip_path)
+    shutil.make_archive(zip_path.replace(".zip", ""), "zip", "./resource")
+    return zip_path
+
+
+custom_css = """
+body {
+    background-color: #eef2f7;
+    margin: 0;
+    padding: 0;
+    font-family: "Microsoft YaHei", sans-serif;
+}
+
+#title-area {
+    text-align: center;
+    margin-top: 30px;
+    margin-bottom: 15px;
+    color: #333;
+}
+
+#chatbot-container, #references-container, #download-container {
+    max-width: 1100px;
+    margin: auto;
+    background: #f0f7ff;
+}
+
+#chatbot-container {
+    border-radius: 12px;
+    box-shadow: 0 4px 10px rgba(0, 0, 0, 0.1);
+    padding: 10px;
+}
+
+#references-container {
+    border-radius: 12px;
+    box-shadow: 0 4px 10px rgba(0, 0, 0, 0.1);
+    padding: 15px;
+}
+
+#submit-btn button {
+    background-color: #007bff;
+    color: white;
+    border-radius: 8px;
+    padding: 8px 12px;
+}
+
+#update-btn button {
+    background-color: #28a745;
+    color: white;
+    border-radius: 8px;
+    padding: 8px 12px;
+}
+
+h3 {
+    color: #444;
+    border-bottom: 2px solid #eee;
+    padding-bottom: 8px;
+    margin-bottom: 20px;
+}
+"""
+custom_head = """
+<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/font-awesome@4.7.0/css/font-awesome.min.css">
+<script src="https://unpkg.com/@dotlottie/player-component@2.7.12/dist/dotlottie-player.mjs" type="module"></script>
+"""
+
+js_func = """
+function refresh() {
+    const url = new URL(window.location);
+
+    if (url.searchParams.get('__theme') !== 'light') {
+        url.searchParams.set('__theme', 'light');
+        window.location.href = url.href;
+    }
+}
+"""
+with gr.Blocks(title="勉強会", css=custom_css, head=custom_head, js=js_func) as demo:
+
+    with gr.Column(elem_id="title-area"):
+        gr.HTML(
+            """
+            <div style="display: flex; align-items: center; justify-content: center;">
+                <img src="https://cdn.profile-image.st-hatena.com/users/saratoga623/profile.png?1728512391" style="width:50px; height:50px; margin-right:10px;">
+                <h1 style="margin: 0;">
+                    <a href="https://saratoga623.hatenablog.com/" target="_blank" style="text-decoration: none; color: inherit;">
+                        プロジェクトマネジメント勉強会 <i class='fa fa-lightbulb-o'></i>
+                    </a>
+                </h1>
+            </div>
+            """
+        )
+        gr.HTML(
+            """
+            <div style="display: flex; align-items: center; justify-content: center;">
+                <div>
+                    <p>経験上、プロジェクトマネジメントで大切だと思っている考え方で</p>
+                    <p>疑問に答え、音声、参考資料が提供されます</p>
+                </div>
+                <dotlottie-player src="https://lottie.host/e2d675e6-73f4-4e92-a217-89bf1a4e49d3/cyif3Z9bwS.lottie"
+                    background="transparent" speed="1" style="width: 150px; height: 150px" loop autoplay></dotlottie-player>
+            </div>
+            """
+        )
+
+    with gr.Column(elem_id="chatbot-container"):
+        gr.HTML("<h3><i class='fa fa-question-circle'></i> 何かお困りですか？</h3>")
+        chatbot = gr.Chatbot(
+            show_label=False,
+            type="messages",
+            bubble_full_width=False,
+            avatar_images=(
+                None,
+                "https://cdn.profile-image.st-hatena.com/users/saratoga623/profile.png?1728512391",
+            ),
+        )
+
+        with gr.Row(elem_id="submit-btn"):
+            msg = gr.Textbox(
+                placeholder="質問を入力してEnter...",
+                show_label=False,
+                scale=5,
+                container=False,
+            )
+            submit_btn = gr.Button(value="送信 🚀", scale=1)
+
+            clear_btn = gr.ClearButton([msg, chatbot], value="クリア 🧹", scale=1)
+
+    with gr.Accordion("ナレッジ一覧", elem_id="references-container", open=False):
+        with gr.Column():
+            with gr.Row():
+                gr.HTML("<h3><i class='fa fa-book'></i> ナレッジ一覧</h3>")
+                update_btn = gr.Button(value="🔄 更新", elem_id="update-btn")
+            references_dropdown = gr.Dropdown(
+                label="ナレッジを選ぶ",
+                choices=list(knowledge_cls.reference_dict.keys()),
+                value=None,
+            )
+            reference_audio = gr.Audio(
+                label="ナレッジ要約を聞く",
+                type="filepath",
+                interactive=False,
+                waveform_options=gr.WaveformOptions(
+                    show_recording_waveform=False,
+                    show_controls=False,
+                ),
+            )
+            sum_references_box = gr.Textbox(
+                label="ナレッジ要約",
+                lines=4,
+                interactive=False,
+            )
+            with gr.Accordion("ナレッジ详情", open=False):
+                all_references_box = gr.Markdown(max_height=250)
+
+    with gr.Accordion("資源", open=False, elem_id="download-container"):
+        download_btn = gr.Button("ダウンロード", elem_id="update-btn")
+        file_output = gr.File(show_label=False, container=False)
+
+    msg.submit(
+        fn=user_message_fn, inputs=[msg, chatbot], outputs=[msg, chatbot], queue=False
+    ).then(fn=answer_question, inputs=chatbot, outputs=chatbot)
+
+    submit_btn.click(
+        fn=user_message_fn, inputs=[msg, chatbot], outputs=[msg, chatbot], queue=False
+    ).then(fn=answer_question, inputs=chatbot, outputs=chatbot)
+    chatbot.like(handle_feedback, None, None, like_user_message=True)
+    update_btn.click(
+        fn=update_knowledge,
+        outputs=references_dropdown,
+    )
+
+    references_dropdown.change(
+        fn=get_reference_info,
+        inputs=references_dropdown,
+        outputs=[reference_audio, all_references_box, sum_references_box],
+    )
+
+    download_btn.click(fn=zip_directory, outputs=file_output)
+
+# app
+if __name__ == "__main__":
+    demo.launch(inline=False, share=False, debug=True)

blog_class.py

@@ -0,0 +1,452 @@
+import os
+import pickle
+import re
+import time
+from datetime import datetime
+from urllib.parse import urljoin
+
+import numpy as np
+import requests
+from bs4 import BeautifulSoup
+from fish_audio_sdk import Session, TTSRequest
+from openai import OpenAI
+from tenacity import retry, stop_after_attempt, wait_exponential
+
+# params
+PKL_FILE = "./resource/knowledge_data.pkl"
+INFO_audio_ID = os.getenv("INFO_audio_ID")
+BASE_URL = "https://saratoga623.hatenablog.com/"
+
+client = OpenAI(api_key=os.getenv("gpt"))
+
+audio_client = Session(os.getenv("audio"))
+
+SYS_Prompt = """
+あなたは言語の専門家であり、さまざまな話し方や思考パターンを分析し、模倣し、適応することに長けています。
+"""
+
+SUMMARY_Prompt = """
+あなたは文章の内容を分析する専門家です。
+以下の文章の要約を作成してください。要約は100字以内で簡潔にまとめてください。
+"""
+
+STYLE_Prompt = """
+あなたは文章の内容を分析する専門家です。
+以下の文章内容を分析して、
+1. 原文の作者の文体を抽出してください。例：敬語、語気、構造、用語など代表的なもの
+2. 原文の作者の思考パターンを抽出してください。例：論理構成、推論、態度など代表的なもの
+"""
+SEGMENT_Prompt = """
+あなたは文章の内容を分析する専門家です。
+以下の文章を内容に基づいて意味的に段落に分割してください。各段落は独立した文章で、段落ごとに改行を入れてください。
+"""
+
+QA_Prompt = """
+ユーザーの質問に応じて表現を的確に調整し、自然でスムーズな対話を実現し、\
+検索された情報の内容、文脈、コミュニケーションスタイルに適した応答を簡潔に提供してください。
+
+【ユーザーの質問】
+{q_text}
+【検索された情報】
+{r_text}
+【タスク要求】
+- 検索された情報を参考に、質問に会話の形で回答すること。
+- 原文の作者の文体を模倣すること。
+- 原文の思考パターンを模倣すること。
+- 簡潔な回答を心がけること。3-5文程度で要点をまとめることを意識してください。
+- 文章風ではなく、対話形式で答えること。
+"""
+REWRITE_SYS_Prompt = """
+Given a conversation (between Human and Assistant) and a follow up message from Human, \
+rewrite the message to be a standalone question that captures all relevant context \
+from the conversation.
+"""
+REWRITE_Prompt = """
+<Chat History>
+{chat_history}
+
+<Follow Up Message>
+{question}
+
+<Standalone question>
+"""
+QA_chat_Prompt = """
+ユーザーの最新質問に応じて表現を的確に調整し、自然でスムーズな対話を実現し、\
+検索された情報の内容、文脈、コミュニケーションスタイルに適した応答を簡潔に提供してください。
+【ユーザーの会話履歴】
+{h_text}
+【ユーザーの最新質問】
+{q_text}
+【検索された情報】
+{r_text}
+【タスク要求】
+- 検索された情報内容を参考に、質問に会話の形で回答すること。
+- 原文の作者の文体を模倣すること。
+- 原文の作者の思考パターンを模倣すること。
+- 簡潔な回答を心がけること。3-5文程度で要点をまとめることを意識してください。
+- 文章風ではなく、対話形式で答えること。
+"""
+
+
+def load_pkl(file_path):
+    if os.path.exists(file_path):
+        with open(file_path, "rb") as f:
+            return pickle.load(f)
+    else:
+        return {}
+
+
+def save_pkl(file_name, data):
+    with open(file_name, "wb") as f:
+        pickle.dump(data, f)
+
+
+def get_page_content(page_url):
+    response = requests.get(page_url)
+    response.encoding = response.apparent_encoding
+    return response.text
+
+
+def parse_homepage(html):
+
+    soup = BeautifulSoup(html, "html.parser")
+    articles = soup.find_all("article", class_="entry")
+    blog_infos = []
+    for article in articles:
+        title_tag = article.find("h1", class_="entry-title")
+        if title_tag and title_tag.find("a"):
+            link = urljoin(BASE_URL, title_tag.find("a")["href"])
+        else:
+            continue
+        # <time> タグ内から公開日を取得する
+        time_tag = article.find("time")
+        if time_tag:
+            year_tag = time_tag.find("span", class_="date-year")
+            month_tag = time_tag.find("span", class_="date-month")
+            day_tag = time_tag.find("span", class_="date-day")
+            if year_tag and month_tag and day_tag:
+                pub_date = f"{year_tag.get_text(strip=True)}-{month_tag.get_text(strip=True)}-{day_tag.get_text(strip=True)}"
+            else:
+                pub_date = "unknown_date"
+        else:
+            pub_date = "unknown_date"
+        blog_infos.append({"date": pub_date, "link": link})
+    return blog_infos, soup
+
+
+def get_next_page_url(soup):
+    next_page_tag = soup.find("a", string="次のページ")
+    if next_page_tag and next_page_tag.has_attr("href"):
+        return urljoin(BASE_URL, next_page_tag["href"])
+    return None
+
+
+def sort_key(info):
+    try:
+        return datetime.strptime(info["date"], "%Y-%m-%d")
+    except Exception:
+        return datetime.max
+
+
+def fetch_blog_text(url):
+    response = requests.get(url)
+    response.encoding = response.apparent_encoding
+    soup = BeautifulSoup(response.text, "html.parser")
+
+    title_tag = soup.find("h1", class_="entry-title")
+    title = title_tag.get_text(strip=True) if title_tag else "no_title"
+
+    content_tag = soup.find("div", class_="entry-content")
+    if not content_tag:
+        content_tag = soup.find("div", class_="hatenablog-entry")
+    if content_tag:
+
+        for a in content_tag.find_all("a", class_="keyword"):
+            a.unwrap()
+        content = content_tag.get_text(strip=False).strip()
+    else:
+        content = ""
+    return title, content
+
+
+def sanitize_filename(filename):
+    return re.sub(r'[\\/*?:"<>|]', "", filename)
+
+
+def max_cosine_similarity(v1, v2_list):
+    """ """
+    v1 = np.array(v1)
+    v2_list = np.array(v2_list)
+
+    norm_v1 = np.linalg.norm(v1)
+    norm_v2 = np.linalg.norm(v2_list, axis=1)
+
+    if norm_v1 == 0:
+        return 1.0 if np.any(norm_v2 == 0) else 0.0
+
+    valid_indices = norm_v2 > 0
+    similarities = np.full(v2_list.shape[0], -1.0)
+    if np.any(valid_indices):
+        similarities[valid_indices] = np.dot(v2_list[valid_indices], v1) / (
+            norm_v1 * norm_v2[valid_indices]
+        )
+    return np.max(similarities)
+
+
+@retry(
+    stop=stop_after_attempt(5),
+    wait=wait_exponential(multiplier=1, min=2, max=10),
+)
+def generate_item(prompt, sys_prompt, model="gpt-4o-mini"):
+    response = client.chat.completions.create(
+        model=model,
+        messages=[
+            {
+                "role": "system",
+                "content": sys_prompt,
+            },
+            {"role": "user", "content": prompt},
+        ],
+    )
+    return response.choices[0].message.content.strip()
+
+
+@retry(
+    stop=stop_after_attempt(5),
+    wait=wait_exponential(multiplier=1, min=2, max=10),
+)
+def get_embedding(text):
+    response = client.embeddings.create(input=text, model="text-embedding-3-small")
+    return [emb.embedding for emb in response.data]
+
+
+@retry(
+    stop=stop_after_attempt(5),
+    wait=wait_exponential(multiplier=1, min=2, max=10),
+)
+def audio_text(text, audio_path):
+    with open(audio_path, "wb") as f:
+        for chunk in audio_client.tts(
+            TTSRequest(reference_id=INFO_audio_ID, text=text)
+        ):
+            f.write(chunk)
+    return audio_path
+
+
+def save_feedback(value, liked):
+    if liked:
+        md_text = "text:\n" + value + "\n" + "liked"
+    else:
+        md_text = "text:\n" + value + "\n" + "disliked"
+    timestamp = int(time.time() * 1000)
+    md_filename = f"./resource/feedback_Text_{timestamp}.md"
+    with open(md_filename, "w", encoding="utf-8") as file:
+        file.write(md_text)
+
+
+class knowledge_class:
+    def __init__(self):
+        self.knowledge_data = load_pkl(PKL_FILE)
+        self.reference_dict = self.get_reference_dict()
+        # q_v = self.knowledge_data["2024-10-09-プロジェクト計画で重要視すること"][
+        #     "vector"
+        # ][0]
+        # t_v = self.knowledge_data["2024-10-09-プロジェクト計画で重要視すること"][
+        #     "vector"
+        # ]
+
+        # cos_sim = max_cosine_similarity(q_v, t_v)
+        # print(cos_sim)
+
+    def get_reference_dict(self):
+        reference_dict = {}
+        for ref_name, ref_dict in self.knowledge_data.items():
+            title = ref_dict["title"]
+            if ref_name not in reference_dict:
+                reference_dict[ref_name] = {}
+                reference_dict[ref_name]["original_text"] = (
+                    f"### {title}\n"
+                    + ref_dict["text"]
+                    + f"\n\n[URL]({ref_dict['url']})"
+                )
+                reference_dict[ref_name]["summary"] = ref_dict["summary"]
+                reference_dict[ref_name]["audio"] = ref_dict["audio"]
+            else:
+                print(f"overlap ref_name: {ref_name}")
+                raise
+        return reference_dict
+
+    def get_new_knowledge(self):
+        print(f"[{datetime.now()}] 記事の検出を開始します...")
+        processed_urls = set()
+        for _, v in self.knowledge_data.items():
+            processed_urls.add(v.get("url", ""))
+        new_blog_infos = []
+        page_url = BASE_URL
+        while page_url:
+            try:
+                html = get_page_content(page_url)
+            except Exception as e:
+                print(f"{page_url} の取得時にエラーが発生しました: {e}")
+                break
+            infos, soup = parse_homepage(html)
+            new_infos = [info for info in infos if info["link"] not in processed_urls]
+            new_blog_infos.extend(new_infos)
+            page_url = get_next_page_url(soup)
+
+        if not new_blog_infos:
+            print("新しい記事は見つかりませんでした。")
+        else:
+            print(
+                f"新規記事 {len(new_blog_infos)} 件を検出しました。処理を開始します..."
+            )
+            new_blog_infos.sort(key=sort_key)
+
+            for info in new_blog_infos:
+                blog_url = info["link"]
+                pub_date = info["date"]
+                print(f"記事を処理中: {blog_url}")
+                try:
+                    title, content = fetch_blog_text(blog_url)
+                    key_name = f"{pub_date}-{sanitize_filename(title)}"
+                    print("記事:", key_name)
+                    summary_text = generate_item(content, SUMMARY_Prompt)
+                    style_text = generate_item(content, STYLE_Prompt)
+                    segment_texts = [
+                        seg.strip()
+                        for seg in generate_item(content, SEGMENT_Prompt).split("\n\n")
+                        if seg.strip()
+                    ]
+                    texts_vector = get_embedding([title, summary_text] + segment_texts)
+                    audio_path = f"./resource/{key_name}.mp3"
+                    audio_path = audio_text(summary_text, audio_path)
+                    dict_item = {
+                        key_name: {
+                            "title": title,
+                            "text": content,
+                            "url": blog_url,
+                            "style": style_text,
+                            "summary": summary_text,
+                            "audio": audio_path,
+                            "segments": segment_texts,
+                            "vector": texts_vector,
+                        }
+                    }
+
+                except Exception as e:
+                    print(f"{blog_url} の処理中にエラーが発生しました: {e}")
+                    raise
+
+                self.knowledge_data.update(dict_item)
+                time.sleep(1)
+                save_pkl(PKL_FILE, self.knowledge_data)
+                time.sleep(1)
+                self.knowledge_data = load_pkl(PKL_FILE)
+                self.reference_dict = self.get_reference_dict()
+                time.sleep(1)
+
+        print(f"PKLファイルの更新が完了しました。新規記事数: {len(new_blog_infos)}")
+
+    def find_top_info(self, question_vector):
+        results = []
+        for idx, k_dict in self.knowledge_data.items():
+            idx_vector = k_dict.get("vector", [0])
+            cos_sim = max_cosine_similarity(question_vector, idx_vector)
+            results.append((idx, cos_sim))
+
+        results_sorted = sorted(results, key=lambda x: x[1], reverse=True)
+        top2 = []
+        retrieve_text = "\n"
+        retrieve_title = ""
+        for res_sort in results_sorted:
+            if res_sort[0] not in top2:
+                top2.append(res_sort[0])
+                retrieve_text += f"情報 {len(top2)} : \n"
+                retrieve_text += (
+                    f"- タイトル：{self.knowledge_data[res_sort[0]]['title']}  \n"
+                )
+                retrieve_title += f"{len(top2)}. {res_sort[0]}  \n"
+                retrieve_text += (
+                    f"- コンテンツ：\n {self.knowledge_data[res_sort[0]]['text']} \n"
+                )
+                retrieve_text += f"- 作者の文体と思考パターン：\n {self.knowledge_data[res_sort[0]]['style']} \n"
+                retrieve_text += f"- 質問と類似度：{res_sort[1]} \n"
+                retrieve_text += "\n"
+            if len(top2) > 1:
+                break
+        return retrieve_text, retrieve_title
+
+    def get_answer(self, question_text):
+
+        # get similar info
+        question_vector = get_embedding(question_text)[0]
+        info_text, info_title = self.find_top_info(question_vector)
+        user_prompt = QA_Prompt.format(q_text=question_text, r_text=info_text)
+        answer_text = generate_item(user_prompt, SYS_Prompt, model="gpt-4o")
+        md_text = user_prompt + "\n 応答: \n" + answer_text
+        timestamp = int(time.time() * 1000)
+        md_filename = f"./resource/QA_Text_{timestamp}.md"
+        audio_filename = f"./resource/QA_Audio_{timestamp}.mp3"
+        with open(md_filename, "w", encoding="utf-8") as file:
+            file.write(md_text)
+        return answer_text, info_title, audio_filename
+
+    def get_chat_answer(self, chat_list):
+        print("all chat list:", chat_list)
+        chat_history = []
+        for turn in chat_list:
+            if turn["role"] == "assistant" and turn["metadata"]:
+                continue
+            else:
+                chat_history.append(turn)
+        chat_history = chat_history[-5:]
+        if chat_history[-1]["role"] != "user" or (not chat_history[-1]["content"]):
+            return None, None, None
+        if len(chat_history) == 1:
+            answer_text, info_title, audio_filename = self.get_answer(
+                chat_history[-1]["content"]
+            )
+            return answer_text, info_title, audio_filename
+        # get history
+        chat_history_str = ""
+        for msg in chat_history[:-1]:
+            chat_history_str += (
+                "\n" if chat_history_str else ""
+            ) + f"{msg['role']}: {msg['content']}"
+        print("chat_history_str:", chat_history_str)
+        new_query = chat_history[-1]["content"]
+        print("new_query:", new_query)
+
+        rw_prompt = REWRITE_Prompt.format(
+            chat_history=chat_history_str, question=new_query
+        )
+
+        if "test" in chat_history[-1]["content"]:
+            return (
+                f"test_ui:{chat_history[-1]['content']}",
+                "1. info_text1 \n 2. info_text2  \n",
+                "./resource/2025-03-03-機嫌良く働くと仕事は上手く進む.mp3",
+            )
+
+        rewirte_question = generate_item(rw_prompt, REWRITE_SYS_Prompt, model="gpt-4o")
+        print("rewirte_question:", rewirte_question)
+        # prompt = DEFAULT_TEMPLATE.format(chat_history=chat_history, question=query)
+        # get rewrite question
+        # get similar info
+        question_vector = get_embedding(rewirte_question)[0]
+        info_text, info_title = self.find_top_info(question_vector)
+        user_prompt = QA_chat_Prompt.format(
+            h_text=chat_history_str, q_text=rewirte_question, r_text=info_text
+        )
+        answer_text = generate_item(user_prompt, SYS_Prompt, model="gpt-4o")
+        md_text = user_prompt + "\n 応答: \n" + answer_text
+        timestamp = int(time.time() * 1000)
+        md_filename = f"./resource/QA_Text_{timestamp}.md"
+        audio_filename = f"./resource/QA_Audio_{timestamp}.mp3"
+        with open(md_filename, "w", encoding="utf-8") as file:
+            file.write(md_text)
+        return answer_text, info_title, audio_filename
+
+
+# kc_class = knowledge_class()
+# print(kc_class.get_new_knowledge())

requirements.txt

@@ -0,0 +1,6 @@
+gradio
+openai
+# zyphra
+tenacity
+beautifulsoup4
+fish-audio-sdk

resource/2024-10-09-プロジェクト計画で重要視すること.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7683d2a43cbeac7888aa3c0e7ffde46be1c358a5570e67835ed54b7431c35f7
+size 387866

resource/2024-10-10-プロジェクト計画時点でユーザーと握ること.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:877e099d232c1b65ccdc5bad4f341ae2c8106cec8d62d063caccab123f44c02d
+size 459755

resource/2024-10-11-全員が同じベクトルを向くために.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26b683d71496c8bfd82df704f1573acd1bb4d65c85896d9484d64ae41bbbbd18
+size 427990

resource/2024-10-12-PMはバランスを取る仕事.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a0f41c346ab19723446bf782c4c49729b17993a460d6c74aeab94cde260c761
+size 478981

resource/2024-10-15-PMはコミュニケーションが大事です.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0df2921d026d1a359762a5066c65b2767068ef34858c0055cad1d77872cdaf92
+size 451813

resource/2024-10-17-進捗報告は定量的に.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60b350f070de3f590c296e3ff6080c138936ef53a3226fc20cd6df77401b7216
+size 450142

resource/2024-10-18-進捗報告から真実を掴むには.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:689b1a8971c2323e9dcc9623c4424203b3558917a0ef6b343f85e1b6cbb6faf7
+size 469368

resource/2024-10-19-報告は多少疑うべし.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6338670de9426f6c652ad97156b92cc58af712a3dfc2ead2375991d187f32f4
+size 462680

resource/2024-10-21-心配事がリスクです.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7df6358c906ff9feddae99f2c082401c7705ab96bb6d30d15393577e7c0d106
+size 351921

resource/2024-10-22-適切なリスク計上のコツ.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa7113ac3ab1528cab5ea2f3f485dd13c641a77b4bf3ad7b82238c45a4d9ca3c
+size 363206

resource/2024-10-23-確実にリスク対策するために.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07bf318bd874b4a1ce3391b1a5acc71ba69ecef4546016648b07397be7a32901
+size 340218

resource/2024-10-24-リスクが顕在化した時の為にやっておくこと.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9929731d716d857f588f7ead9d94260329b0749026002dd9a5fe220a228455d
+size 418377

resource/2024-10-25-前提をリスクで管理する.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:021a07cc4025677192403f415f88c50317baa30d4e0539f364e98cba7e1f2b63
+size 447216

resource/2024-10-26-言い訳をリスクで管理する.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ac122cd688d79ad7e3cf6569b8fd156e2ad2352fed1ac7871efe0fc8fa7ef4c
+size 332695

resource/2024-10-27-リスクを棚卸しする.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0380b69bb7802ff46f02f243c0474a5c13623246074a5cfb6ec3f0e22299e68
+size 359444

resource/2024-10-28-課題の対策を考える.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d56a6f13943fdccbb953427737f9d7da4cfced3f1b782259dcdefb111a70ed0a
+size 393717

resource/2024-10-29-課題の早期発見に向けて心がけること.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a55eaf799452cd8eeb9d4432495f978ed794b18a4836a75a4ebeb03b2d30f23
+size 443454

resource/2024-10-30-現実的な対策を考える.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02e32622eec4b6e8aa96ffe7d1a0a810594b6f5fcd573d0ed89ce7b7fd0cbc47
+size 532062

resource/2024-10-31-報告の目的って何？.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:413fb10503a6ee67db413a0f9524c2c325794ecda741d15e5b47338915a099dc
+size 566752

resource/2024-11-01-原因と真因.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d600f5bc08799134326da48bc54bb9641779bec47e529d9aa8bdf1517b6ed57e
+size 354847

resource/2024-11-02-なぜを5回繰り返してもダメな理由.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c47bbc5846a701eb303b8839cb228d31a9485512a8e92f996fed12294cee8be
+size 341472

resource/2024-11-03-真因分析の極意.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ceea59afe99cb1d233f2cc5a35da9d5e98c0b70b82b14d217217559f7056e5a
+size 341054

resource/2024-11-05-対策検討の極意.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7aa29e7c6c92e693f679a84db7be50e318774cf65c0c920a52fb440d1f2bde90
+size 371565

resource/2024-11-06-プロジェクトは性悪説がちょうどいい.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4866f65c0fc02893caad25e540789b76e3d3c43793a4c0ea969c401132cdddb5
+size 432169

resource/2024-11-07-コスト管理でよくある話.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b2aeb9e6206267a90b8c4b3c65d3d5a3ccd42b31c24f41bb3df50e2390a2400
+size 417541

resource/2024-11-08-手戻りはもったいない.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3319adaac8419d57b1ea92675c7443bfaec10ffb250483a73ea05bc1d23268b9
+size 363206

resource/2024-11-09-責任って何？.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:591c5ddf4ccda6df6274a1d61279b99f9967ffa51714a37ddf16d5c32cb7cbdd
+size 419631

resource/2024-11-10-早く開発するためには.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:751232225cf687bd6643c27d37f059606c8ba03c6ea810eb7a3a39cdbb9deac3
+size 383268

resource/2024-11-11-変更管理の要諦.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:814aa9ef7503e9397aafcd796a607093c5c47fb4a1412b165920614e34435974
+size 419631

resource/2024-11-12-立ち止まる勇気.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35f27fb560a5a68361baac4bd8a6edb86dc457443193b058c876403cc4fbd652
+size 429244

resource/2024-11-13-テストの目的.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:378667439071712bdbc6311e1e1733239b903990a4c46d3116ef0e28f5c763df
+size 413779

resource/2024-11-14-テスト計画のコツ.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d87dfecca48dddfd14770cca39e79375056381a06fb1d2006c3309b205a92886
+size 349831

resource/2024-11-15-雛形の必要性と弊害.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a267e3b27399e305fa59823ef1fa3e1ba6c66e3cb6a3d3893484c5ed5b8ab59
+size 533315

resource/2024-11-16-個人を主語にするステークホルダーマネジメント.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5e283edea25048840152854628bf9df614b4c4d2785cbdb63be2ce018bfd42e
+size 381596

resource/2024-11-17-議論する組織を目指す.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e778c62045eccf8a523d32368830c06f425a3ec1cac67aecaf2fa38be0f6061
+size 398733

resource/2024-11-18-問題を問題と言える風土.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6cc92081b4d3fd42c206781043bb31d2a42afdba01c656c42fb0d1c217d8f05f
+size 345234

resource/2024-11-19-保守工程のことを意識する.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d44f8eb264c287704c2058b9328846b058aa033e5e0bae9c71b9ae17c67eec56
+size 401658

resource/2024-11-20-本番実施計画も性悪説が基本です.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b541fbaf1989b79d7b0b674a4b9ecde49e5d254968e02d9b6ff0bbca40ee612c
+size 406256

resource/2024-11-21-本番確認のレベル感を決める観点.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ebef11621b3b80e06912022d42b9f45355cccb11670bb9d9b715a9fa782f3a0
+size 476055

resource/2024-11-22-失敗のパターンを持っておく.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52f2e0a8c464cc7a2d35c765c08352c8e7cbee964b14cf8187fb0938fd45e741
+size 337293

resource/2024-11-23-規模の小さいお仕事で基礎・基本を身につける.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:925c2514fffcde5b64c553dbf992d6b36a8f4457bfbf77499912999f877abfd4
+size 361116

resource/2024-11-24-極力マルチタスクを作らない計画.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4da8741bf47da797a160af970d67b5c4a8177f900884cf81ffbd2b3836395bf9
+size 473965

resource/2024-11-25-「頑張ります」を適切に使い、正しく解釈する.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67940e9700a5bb82618b9def888d724da18aefa1eb4c58a113f3b03aba425ca2
+size 478981

resource/2024-11-26-基本設計の品質がプロジェクト成功の鍵.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e6cf764ff4449a6be3a33e6e0c92f3f2dfaa069c068e0f71ad8f11adef60935
+size 380342

resource/2024-11-27-文章を使ってプロジェクトを成功に導く.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a0eb4054fbfc33baee0ea523e94aaaba5560a93d7e1c19d613983f3bc83bdd7
+size 489430

resource/2024-11-28-作戦を伝えることがプロジェクトをスムーズに進める鍵.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c628fd8f82c01506f61e86d36d6c2a3c4a4d300a320038076da5856f8ababb12
+size 438439