import json import time from typing import Dict, Any import gradio as gr # --- Simulation Setup for LLM API --- # This section simulates the core AI generation logic without requiring a live API key. LLM_MODEL = "gemini-2.5-flash-preview-09-2025" API_KEY = "" # API Key Placeholder def simulate_gemini_api_call(payload: Dict[str, Any], fields: Dict[str, Any]) -> Dict[str, Any]: """ Simulates a structured response from the Gemini API based on the task type. In a real application, this function would make a fetch call to the Gemini API. """ # Simulate API latency time.sleep(1.0) user_query = payload['contents'][0]['parts'][0]['text'] system_instruction = payload.get('systemInstruction', {}).get('parts', [{}])[0].get('text', 'No system instruction') # Check system instruction to determine the output type (Admin or Teaching) if "台灣中學學務處行政書記" in system_instruction: # Simulate Admin Copilot (Meeting Minutes) output mock_text_result = json.dumps({ "文件類型 (Document Type)": "學務處會議記錄 (Academic Affairs Meeting Minutes)", "meeting_info": { "date": fields.get('date', '2025-01-10'), "location": fields.get('location', '學務處會議室'), "topic": fields.get('topic', '模擬會議主題') }, "attendees": ["校長", "學務主任", "衛生組長", "生輔組長"], "key_points": [ "期末獎懲核定程序已完成，共核定 30 件。建議將名單呈報校長核閱。", "新生訓練場地佈置進度達 80%，物資清單已交付總務處採購。", f"重點輸入: {fields.get('key_input', 'N/A')}" ], "resolutions": [ {"item": "發布正式期末獎懲公告。", "responsible": "教務處", "deadline": "2025-01-15"}, {"item": "新生訓練場地佈置於活動前一天完成驗收。", "responsible": "總務處", "deadline": "2025-08-20"} ], "audit_note": "文件根據校內行政公文標準格式生成。" }, ensure_ascii=False, indent=2) elif "台灣國高中資深教師與課程設計師" in system_instruction: # Simulate Teaching Designer (Lesson Plan & Rubric) output mock_text_result = json.dumps({ "文件類型 (Document Type)": "單元教案與評量規準 (Lesson Plan & Rubric)", "lesson_plan_title": f"【{fields.get('subject', 'N/A')}】探索 {fields.get('topic', 'N/A')} ({fields.get('hours', 0)} 課時)", "grade_level": fields.get('grade', 'N/A'), "curriculum_alignment": ["A2 邏輯推理與批判思辨", "B3 獨立思考與探究精神"], "learning_objectives": ["學生能解釋核心概念 X。", "學生能應用方法 Y 進行分析。", "學生能製作報告Z進行表達。"], "activities": [ {"time_min": 15, "stage": "引導", "method": "提問式教學", "description": "使用新聞案例引導核心概念。"}, {"time_min": 30, "stage": "活動一", "method": "合作學習", "description": "分組完成專題研究和實作練習。"}, ], "rubric": { "title": "單元評量規準 (4 級 X 4 指標)", "criteria": [ {"name": "概念理解", "A": "清晰精確地解釋所有核心概念。", "D": "只能回答簡單問題。"}, {"name": "協作能力", "A": "積極領導團隊完成任務。", "D": "未參與討論。"} ] }, "differentiation_advice": f"根據班級特性 ({fields.get('class_needs', 'N/A')})，建議提供圖像化教材並進行分組輔導。" }, ensure_ascii=False, indent=2) else: mock_text_result = json.dumps({"error": "Unknown or missing task instruction."}) # Return the simulated API response structure return { "candidates": [{ "content": { "parts": [{ "text": mock_text_result }] }, "groundingMetadata": {} }] } # --- Module A: Admin Copilot Generator (Gradio Wrapper) --- def admin_copilot_generator(template_id: str, topic: str, date: str, location: str, key_input: str) -> str: """ Handles the Admin Copilot UI inputs and calls the simulation. """ fields = { "topic": topic, "date": date, "location": location, "key_input": key_input } # System Prompt defined for the Admin Copilot system_prompt = ( "角色：台灣中學學務處行政書記\n" "輸出：JSON（會議資訊、出席、重點、決議、待辦、負責人、期限）\n" "格式規範：用詞正式、避免口語、保留專有名詞\n" "限制：所有決議必須有負責人和明確期限。" ) # Response Schema is implicitly defined but would be included in a real API call. # The Gradio JSON output will just display the resulting JSON string. user_query = f"請生成一份會議記錄。主題: {topic}; 輸入重點（或逐字稿）：{key_input}" payload = { "contents": [{ "parts": [{ "text": user_query }] }], "systemInstruction": { "parts": [{ "text": system_prompt }] }, # Simplified generationConfig for simulation "generationConfig": { "responseMimeType": "application/json" } } api_response = simulate_gemini_api_call(payload, fields) try: json_string = api_response['candidates'][0]['content']['parts'][0]['text'] # For Gradio, we return the JSON string directly return json_string except (KeyError, json.JSONDecodeError) as e: return f"ERROR: Failed to parse LLM structured output. {e}" # --- Module B: Teaching AI Designer (Gradio Wrapper) --- def lesson_plan_designer(grade: str, subject: str, topic: str, hours: float, method: str, equipment: str, class_needs: str) -> str: """ Handles the Teaching Designer UI inputs and calls the simulation. Note: hours is float because Gradio Slider output is float """ fields = { "grade": grade, "subject": subject, "topic": topic, "hours": int(hours), # Convert back to int for display consistency "method": method, "equipment": equipment, "class_needs": class_needs } # System Prompt defined for the Teaching Designer system_prompt = ( "角色：台灣國高中資深教師與課程設計師\n" "輸出：JSON（教案標題、目標、課綱對齊、活動步驟、評量規準、差異化建議）\n" "限制：活動分鏡以 15 分鐘粒度；至少 2 項形成性評量。\n" "對齊：請將輸出中的 'curriculum_alignment' 欄位，對齊台灣課綱的關鍵能力/素養。" ) user_query = ( f"請根據以下資訊設計一個單元教案、評量規準和差異化建議：\n" f"年級/學科/單元主題: {grade}/{subject}/{topic}\n" f"課時數: {int(hours)} 節\n" f"教學法偏好: {method}\n" f"可用設備: {equipment}\n" f"班級特性: {class_needs}" ) payload = { "contents": [{ "parts": [{ "text": user_query }] }], "systemInstruction": { "parts": [{ "text": system_prompt }] }, # Simplified generationConfig for simulation "generationConfig": { "responseMimeType": "application/json" } } api_response = simulate_gemini_api_call(payload, fields) try: json_string = api_response['candidates'][0]['content']['parts'][0]['text'] return json_string except (KeyError, json.JSONDecodeError) as e: return f"ERROR: Failed to parse LLM structured output. {e}" # --- Gradio Interface Definition --- # Module A Interface (Admin Copilot) admin_copilot_interface = gr.Interface( fn=admin_copilot_generator, inputs=[ gr.Textbox(label="模板 ID (Template ID - Fixed for MVP)", value="meeting_minutes_standard", interactive=False), gr.Textbox(label="會議主題 (Meeting Topic)", value="學務處期末獎懲與新生訓練籌備會議"), gr.Textbox(label="日期 (Date)", value="2025-01-10"), gr.Textbox(label="地點 (Location)", value="學務處會議室"), gr.Textbox(label="輸入重點/逐字稿 (Key Input/Transcript)", value="討論期末獎懲核定程序。新生訓練場地佈置、人員編組確認。", lines=5), ], outputs=gr.JSON(label="AI 生成結構化 JSON (原始資料)"), title="行政 Copilot：會議記錄生成 (Admin Copilot: Meeting Minutes Generation)", description="🎯 生成格式嚴謹的行政文件 JSON 結構。", flagging_mode="never", # Updated from allow_flagging ) # Module B Interface (Teaching Designer) lesson_plan_designer_interface = gr.Interface( fn=lesson_plan_designer, inputs=[ gr.Dropdown(label="年級 (Grade)", choices=["國中", "高中", "國小"], value="高中"), gr.Textbox(label="學科 (Subject)", value="歷史"), gr.Textbox(label="單元主題 (Unit Topic)", value="從茶葉看全球化：17-19世紀的貿易網絡"), gr.Slider(label="課時數 (Number of Sessions)", minimum=1, maximum=10, step=1, value=4), gr.Dropdown(label="教學法偏好 (Pedagogy Preference)", choices=["探究式、PBL", "翻轉教學", "合作學習", "講述法"], value="探究式、PBL"), gr.Textbox(label="可用設備 (Available Equipment)", value="平板電腦、投影設備、網路"), gr.Textbox(label="班級特性 (Class Characteristics)", value="班級組成多元，需考慮多樣化的史料呈現方式。"), ], outputs=gr.JSON(label="AI 生成教案與評量規準 JSON (原始資料)"), title="教學 AI 設計器：教案與 Rubric 生成 (Teaching AI Designer: Lesson Plan & Rubric)", description="📘 生成符合課綱精神的單元教案結構和評量規準 JSON。", flagging_mode="never", # Updated from allow_flagging ) # Integrate the two modules into a Tabbed Interface demo = gr.TabbedInterface( [admin_copilot_interface, lesson_plan_designer_interface], ["模組 A: 行政 Copilot", "模組 B: 教學設計器"], title="CampusAI Suite (台灣校園 AI 文書/教學 MVP 演示)", theme=gr.themes.Soft() ) # --- Launch the application --- # This is required for the application to start in the container if __name__ == "__main__": demo.launch()