Spaces:

SWYfisher
/

aichatbot

Sleeping

Update app.py

ee614bc verified 6 months ago

15.7 kB

# ==========================================
# MILESTONE 6: PRODUCTION READY PLATFORM
# Multi-Format File Support (.pdf, .txt, .docx, .csv, .md)
# ==========================================

print("Installing required tools...")
# ADDED: docx2txt to read Microsoft Word files

import gradio as gr
from huggingface_hub import InferenceClient
from transformers import pipeline
import numpy as np
import librosa
import traceback
import torch
import os
import shutil
import json
import zipfile

# --- UPGRADED RAG IMPORTS ---
# We imported Docx2txtLoader and CSVLoader!
from langchain_community.document_loaders import PyPDFLoader, TextLoader, Docx2txtLoader, CSVLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS

# ------------------------------------------
# 0. DIRECTORIES & DYNAMIC CONFIGURATION
# ------------------------------------------
DOC_DIR = "school_data"
CONFIG_FILE = os.path.join(DOC_DIR, "config.json")
EXPORT_DIR = "deployment_export"
os.makedirs(DOC_DIR, exist_ok=True)
os.makedirs(EXPORT_DIR, exist_ok=True)

DEFAULT_CONFIG = {
    "llm_model": "Qwen/Qwen2.5-7B-Instruct",
    "prompt_school": "You are a school assistant. Answer using strictly this info:",
    "prompt_general": "You are a friendly general tutor. Answer the question nicely.",
    "max_audio_sec": 15,
    "max_file_mb": 5
}

def load_config():
    if os.path.exists(CONFIG_FILE):
        with open(CONFIG_FILE, 'r') as f: return json.load(f)
    return DEFAULT_CONFIG

def save_config(new_config):
    with open(CONFIG_FILE, 'w') as f: json.dump(new_config, f)

if not os.path.exists(CONFIG_FILE): save_config(DEFAULT_CONFIG)

# ------------------------------------------
# 1. RAG & AI INITIALIZATION (MULTI-FILE LOGIC)
# ------------------------------------------
knowledge_base = None
print("Downloading the Document Reading AI...")
embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")

def rebuild_knowledge_base():
    global knowledge_base
    files = [f for f in os.listdir(DOC_DIR) if f != "config.json" and not f.startswith('.')]
    if len(files) == 0:
        knowledge_base = None
        return "⚠️ All files deleted. AI brain is empty."

    all_chunks = []
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)

    for filename in files:
        filepath = os.path.join(DOC_DIR, filename)
        try:
            # --- THE MULTI-FORMAT ROUTER ---
            if filename.lower().endswith(".pdf"):
                loader = PyPDFLoader(filepath)
            elif filename.lower().endswith(".txt") or filename.lower().endswith(".md"):
                loader = TextLoader(filepath, encoding='utf-8')
            elif filename.lower().endswith(".docx"):
                loader = Docx2txtLoader(filepath)
            elif filename.lower().endswith(".csv"):
                loader = CSVLoader(filepath)
            else:
                print(f"Skipping unsupported file type: {filename}")
                continue

            all_chunks.extend(text_splitter.split_documents(loader.load()))
        except Exception as e:
            print(f"Error reading {filename}: {e}")

    if len(all_chunks) > 0:
        knowledge_base = FAISS.from_documents(all_chunks, embeddings)
        return f"✅ Brain updated! {len(files)} files loaded."
    knowledge_base = None
    return "⚠️ Files found, but no text extracted."

my_secret_key = os.getenv("Token")
client = InferenceClient(token=my_secret_key)
compute_device = 0 if torch.cuda.is_available() else -1
local_whisper = pipeline("automatic-speech-recognition", model="openai/whisper-small", chunk_length_s=30, device=compute_device)
rebuild_knowledge_base()

# ------------------------------------------
# 2. CORE BACKEND LOGIC
# ------------------------------------------
def get_file_list():
    return [f for f in os.listdir(DOC_DIR) if f != "config.json" and not f.startswith('.')]

def upload_files(files):
    if files is not None:
        config = load_config()
        max_bytes = config["max_file_mb"] * 1024 * 1024

        for f in files:
            if os.path.getsize(f.name) > max_bytes:
                gr.Warning(f"Blocked: {os.path.basename(f.name)} exceeds {config['max_file_mb']}MB limit.")
                continue

            filename = os.path.basename(f.name)
            shutil.copy(f.name, os.path.join(DOC_DIR, filename))

    gr.Info(rebuild_knowledge_base())
    return gr.update(choices=get_file_list()), gr.update(value=None)

def delete_selected_files(selected_filenames):
    if selected_filenames:
        for filename in selected_filenames:
            filepath = os.path.join(DOC_DIR, filename)
            if os.path.exists(filepath): os.remove(filepath)
    gr.Info(rebuild_knowledge_base())
    return gr.update(choices=get_file_list(), value=[])

def download_selected_files(selected_filenames):
    if not selected_filenames:
        gr.Warning("Please select files to download first.")
        return gr.update(visible=False, value=None)
    return gr.update(visible=True, value=[os.path.join(DOC_DIR, f) for f in selected_filenames])

def transcribe_only(audio_input, language_choice):
    if audio_input is None: return ""
    try:
        config = load_config()
        original_sr, audio_array = audio_input

        audio_duration = len(audio_array) / original_sr
        if audio_duration > config["max_audio_sec"]:
            gr.Warning(f"Audio blocked! Exceeds the {config['max_audio_sec']} second limit.")
            return ""

        if audio_array.ndim > 1: audio_array = np.mean(audio_array, axis=1)
        audio_array = audio_array.astype(np.float32)
        if np.max(np.abs(audio_array)) > 1.0: audio_array = audio_array / 32768.0
        if original_sr != 16000: audio_array = librosa.resample(y=audio_array, orig_sr=original_sr, target_sr=16000)

        ai_lang = "english" if language_choice == "English" else "chinese"
        result = local_whisper({"sampling_rate": 16000, "raw": audio_array}, generate_kwargs={"language": ai_lang, "task": "transcribe"})
        return result['text']
    except Exception: return ""

def agent_chat(user_text, chat_history):
    global knowledge_base
    try:
        if not user_text or not str(user_text).strip():
            gr.Warning("⚠️ 請輸入文字！")
            return "", chat_history

        config = load_config()
        user_input_str = str(user_text) # 強制轉成字串，防止意外

        # --- 1. 路由判斷 (確保格式正確) ---
        router_prompt = f"Determine if this is SCHOOL or GENERAL: {user_input_str}"
        route_res = client.chat_completion(
            model=config["llm_model"], 
            messages=[{"role": "user", "content": router_prompt}], # 這裡必須是字典
            max_tokens=5
        )

        # --- 2. 取得 System Prompt ---
        route_verdict = route_res.choices[0].message.content.strip()
        if "SCHOOL" in route_verdict:
            if knowledge_base is not None:
                docs = knowledge_base.similarity_search(user_input_str, k=2)
                context = "\n".join([doc.page_content for doc in docs])
                system_prompt = f"{config['prompt_school']}\nContext: {context}"
            else:
                system_prompt = "You are a school assistant. No documents found."
        else:
            system_prompt = config['prompt_general']

        # --- 3. 關鍵修正：構建發送給 AI 的完整清單 ---
        # 我們在這裡明確定義 role 和 content，確保完全符合 API 要求
        final_messages = [
            {"role": "system", "content": str(system_prompt)}
        ]

        # 如果你想加入記憶（History），請用這個迴圈：
        for old_user, old_ai in chat_history[-5:]: # 取最近5次對話
            final_messages.append({"role": "user", "content": str(old_user)})
            final_messages.append({"role": "assistant", "content": str(old_ai)})

        # 加入本次的問題
        final_messages.append({"role": "user", "content": user_input_str})

        # --- 4. 發送請求 ---
        final_response = client.chat_completion(
            model=config["llm_model"],
            messages=final_messages, # 傳入剛才建好的字典清單
            max_tokens=200,
            temperature=0.5
        )

        ai_reply = final_response.choices[0].message.content
        chat_history.append((user_input_str, ai_reply))
        return "", chat_history

    except Exception as e:
        print(f"DEBUG ERROR: {e}") # 在後台印出真正的錯誤原因
        chat_history.append(("Error", f"⚠️ 格式錯誤或連線中斷: {str(e)}"))
        return "", chat_history
# ------------------------------------------
# 3. ADMIN SETTINGS HELPERS
# ------------------------------------------
def admin_login(u, p):
    if u == "admin" and p == "admin": return gr.update(visible=False), gr.update(visible=True)
    return gr.update(visible=True), gr.update(visible=False)

def update_ai_settings(model, p_school, p_general):
    conf = load_config()
    final_model = model.strip() if model and model.strip() != "" else "Qwen/Qwen2.5-7B-Instruct"
    conf.update({"llm_model": final_model, "prompt_school": p_school, "prompt_general": p_general})
    save_config(conf)
    gr.Info(f"✅ AI Config Saved! Model set to: {final_model}")
    return final_model

def update_security_settings(audio_sec, file_mb):
    conf = load_config()
    conf.update({"max_audio_sec": audio_sec, "max_file_mb": file_mb})
    save_config(conf)
    gr.Info("🛡️ Security Guards updated successfully!")

def export_project():
    zip_path = os.path.join(EXPORT_DIR, "school_bot_deploy.zip")
    req_text = "gradio\ntransformers\ntorch\nlibrosa\nsoundfile\ndocx2txt\nlangchain\nlangchain-community\nlangchain-huggingface\nlangchain-text-splitters\nfaiss-cpu\npypdf\nhuggingface_hub"
    with zipfile.ZipFile(zip_path, 'w') as zf:
        zf.writestr("requirements.txt", req_text)
        zf.writestr("README.md", "# Deployment Guide\n1. Upload this code to Hugging Face Spaces.\n2. Add your HF_TOKEN to the Space Secrets.\n3. Run app.py!")
    return gr.update(visible=True, value=zip_path)

# ------------------------------------------
# 4. THE UI LAYOUT
# ------------------------------------------
print("Building Final UI...")
current_config = load_config()

with gr.Blocks(theme=gr.themes.Soft()) as demo:

    with gr.Tab("💬 Student Chatbot"):
        gr.Markdown("# 🦸‍♂️ My Ultimate Hybrid AI Agent")
        chatbot = gr.Chatbot(height=400)
        with gr.Row():
            with gr.Column(scale=3): txt_input = gr.Textbox(label="Type a message")
            with gr.Column(scale=2):
                lang_toggle = gr.Radio(choices=["English", "Chinese"], value="Chinese", label="Voice Language:")
                aud_input = gr.Audio(sources=["microphone"], type="numpy", label="Record Voice")
        with gr.Row():
            submit_btn = gr.Button("🚀 Send to AI", variant="primary")
            clear_btn = gr.ClearButton([txt_input, aud_input, chatbot], value="🗑️ Clear Chat")

        aud_input.change(fn=transcribe_only, inputs=[aud_input, lang_toggle], outputs=[txt_input])
        submit_btn.click(fn=agent_chat, inputs=[txt_input, chatbot], outputs=[txt_input, chatbot])

    with gr.Tab("⚙️ Admin Settings"):
        with gr.Column(visible=True) as login_screen:
            gr.Markdown("### 🔒 Admin Login")
            with gr.Row():
                usr_box = gr.Textbox(label="Username")
                pwd_box = gr.Textbox(label="Password", type="password")
            login_btn = gr.Button("Login", variant="primary")

        with gr.Column(visible=False) as admin_panel:
            gr.Markdown("## 🎛️ Platform Control Center")

            with gr.Tabs():
                with gr.Tab("📂 Data Manager"):
                    with gr.Row():
                        # THE UI UPGRADE: Now accepts .docx, .csv, and .md!
                        upload_btn = gr.File(
                            label="1. Upload File (Click or Drag & Drop)",
                            file_count="multiple",
                            type="filepath",
                            file_types=[".pdf", ".txt", ".docx", ".csv", ".md"]
                        )
                        with gr.Column():
                            download_btn = gr.Button("2. ⬇️ Download Selected")
                            delete_btn = gr.Button("3. 🗑️ Delete Selected", variant="stop")
                            download_output = gr.File(label="Click to Save", visible=False)
                    gr.Markdown("---")
                    with gr.Row():
                        gr.Markdown("### 📄 Database List View")
                        select_all_btn = gr.Button("☑️ Select All", size="sm")
                    file_listview = gr.CheckboxGroup(choices=get_file_list(), label="")

                with gr.Tab("🧠 AI Config & Prompts"):
                    gr.Markdown("Change how the AI behaves in real-time without restarting the server.")
                    ai_model_txt = gr.Textbox(value=current_config["llm_model"], label="Cloud LLM Model (Hugging Face ID)", placeholder="Leave blank to use default (Qwen/Qwen2.5-7B-Instruct)")
                    prompt_school_txt = gr.TextArea(value=current_config["prompt_school"], label="School RAG System Prompt")
                    prompt_general_txt = gr.TextArea(value=current_config["prompt_general"], label="General Conversation System Prompt")
                    save_ai_btn = gr.Button("💾 Save AI Settings", variant="primary")

                with gr.Tab("🛡️ Security Guards"):
                    gr.Markdown("Protect your server from spam and massive files.")
                    audio_limit = gr.Slider(minimum=5, maximum=60, step=1, value=current_config["max_audio_sec"], label="Max Voice Recording Length (Seconds)")
                    file_limit = gr.Slider(minimum=1, maximum=50, step=1, value=current_config["max_file_mb"], label="Max Upload File Size (MB)")
                    save_sec_btn = gr.Button("💾 Save Security Settings", variant="primary")

                with gr.Tab("☁️ Cloud Deployment"):
                    gr.Markdown("### Ready to go live 24/7?")
                    gr.Markdown("Google Colab goes to sleep when you close your browser. To keep this bot online forever, export your deployment kit and upload it to a free host like **Hugging Face Spaces**.")
                    export_btn = gr.Button("📦 Generate Deployment Kit (ZIP)", variant="primary")
                    export_output = gr.File(label="Download your ZIP", visible=False)

        # WIRING
        login_btn.click(fn=admin_login, inputs=[usr_box, pwd_box], outputs=[login_screen, admin_panel])
        upload_btn.upload(fn=upload_files, inputs=[upload_btn], outputs=[file_listview, upload_btn])
        delete_btn.click(fn=delete_selected_files, inputs=[file_listview], outputs=[file_listview])
        download_btn.click(fn=download_selected_files, inputs=[file_listview], outputs=[download_output])
        def select_all(): return gr.update(value=get_file_list())
        select_all_btn.click(fn=select_all, inputs=[], outputs=[file_listview])
        save_ai_btn.click(fn=update_ai_settings, inputs=[ai_model_txt, prompt_school_txt, prompt_general_txt], outputs=[ai_model_txt])
        save_sec_btn.click(fn=update_security_settings, inputs=[audio_limit, file_limit])
        export_btn.click(fn=export_project, outputs=[export_output])

demo.launch(share=True)

	# ==========================================
	# MILESTONE 6: PRODUCTION READY PLATFORM
	# Multi-Format File Support (.pdf, .txt, .docx, .csv, .md)
	# ==========================================

	print("Installing required tools...")
	# ADDED: docx2txt to read Microsoft Word files

	import gradio as gr
	from huggingface_hub import InferenceClient
	from transformers import pipeline
	import numpy as np
	import librosa
	import traceback
	import torch
	import os
	import shutil
	import json
	import zipfile

	# --- UPGRADED RAG IMPORTS ---
	# We imported Docx2txtLoader and CSVLoader!
	from langchain_community.document_loaders import PyPDFLoader, TextLoader, Docx2txtLoader, CSVLoader
	from langchain_text_splitters import RecursiveCharacterTextSplitter
	from langchain_huggingface import HuggingFaceEmbeddings
	from langchain_community.vectorstores import FAISS

	# ------------------------------------------
	# 0. DIRECTORIES & DYNAMIC CONFIGURATION
	# ------------------------------------------
	DOC_DIR = "school_data"
	CONFIG_FILE = os.path.join(DOC_DIR, "config.json")
	EXPORT_DIR = "deployment_export"
	os.makedirs(DOC_DIR, exist_ok=True)
	os.makedirs(EXPORT_DIR, exist_ok=True)

	DEFAULT_CONFIG = {
	"llm_model": "Qwen/Qwen2.5-7B-Instruct",
	"prompt_school": "You are a school assistant. Answer using strictly this info:",
	"prompt_general": "You are a friendly general tutor. Answer the question nicely.",
	"max_audio_sec": 15,
	"max_file_mb": 5
	}

	def load_config():
	if os.path.exists(CONFIG_FILE):
	with open(CONFIG_FILE, 'r') as f: return json.load(f)
	return DEFAULT_CONFIG

	def save_config(new_config):
	with open(CONFIG_FILE, 'w') as f: json.dump(new_config, f)

	if not os.path.exists(CONFIG_FILE): save_config(DEFAULT_CONFIG)

	# ------------------------------------------
	# 1. RAG & AI INITIALIZATION (MULTI-FILE LOGIC)
	# ------------------------------------------
	knowledge_base = None
	print("Downloading the Document Reading AI...")
	embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")

	def rebuild_knowledge_base():
	global knowledge_base
	files = [f for f in os.listdir(DOC_DIR) if f != "config.json" and not f.startswith('.')]
	if len(files) == 0:
	knowledge_base = None
	return "⚠️ All files deleted. AI brain is empty."

	all_chunks = []
	text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)

	for filename in files:
	filepath = os.path.join(DOC_DIR, filename)
	try:
	# --- THE MULTI-FORMAT ROUTER ---
	if filename.lower().endswith(".pdf"):
	loader = PyPDFLoader(filepath)
	elif filename.lower().endswith(".txt") or filename.lower().endswith(".md"):
	loader = TextLoader(filepath, encoding='utf-8')
	elif filename.lower().endswith(".docx"):
	loader = Docx2txtLoader(filepath)
	elif filename.lower().endswith(".csv"):
	loader = CSVLoader(filepath)
	else:
	print(f"Skipping unsupported file type: {filename}")
	continue

	all_chunks.extend(text_splitter.split_documents(loader.load()))
	except Exception as e:
	print(f"Error reading {filename}: {e}")

	if len(all_chunks) > 0:
	knowledge_base = FAISS.from_documents(all_chunks, embeddings)
	return f"✅ Brain updated! {len(files)} files loaded."
	knowledge_base = None
	return "⚠️ Files found, but no text extracted."

	my_secret_key = os.getenv("Token")
	client = InferenceClient(token=my_secret_key)
	compute_device = 0 if torch.cuda.is_available() else -1
	local_whisper = pipeline("automatic-speech-recognition", model="openai/whisper-small", chunk_length_s=30, device=compute_device)
	rebuild_knowledge_base()

	# ------------------------------------------
	# 2. CORE BACKEND LOGIC
	# ------------------------------------------
	def get_file_list():
	return [f for f in os.listdir(DOC_DIR) if f != "config.json" and not f.startswith('.')]

	def upload_files(files):
	if files is not None:
	config = load_config()
	max_bytes = config["max_file_mb"] * 1024 * 1024

	for f in files:
	if os.path.getsize(f.name) > max_bytes:
	gr.Warning(f"Blocked: {os.path.basename(f.name)} exceeds {config['max_file_mb']}MB limit.")
	continue

	filename = os.path.basename(f.name)
	shutil.copy(f.name, os.path.join(DOC_DIR, filename))

	gr.Info(rebuild_knowledge_base())
	return gr.update(choices=get_file_list()), gr.update(value=None)

	def delete_selected_files(selected_filenames):
	if selected_filenames:
	for filename in selected_filenames:
	filepath = os.path.join(DOC_DIR, filename)
	if os.path.exists(filepath): os.remove(filepath)
	gr.Info(rebuild_knowledge_base())
	return gr.update(choices=get_file_list(), value=[])

	def download_selected_files(selected_filenames):
	if not selected_filenames:
	gr.Warning("Please select files to download first.")
	return gr.update(visible=False, value=None)
	return gr.update(visible=True, value=[os.path.join(DOC_DIR, f) for f in selected_filenames])

	def transcribe_only(audio_input, language_choice):
	if audio_input is None: return ""
	try:
	config = load_config()
	original_sr, audio_array = audio_input

	audio_duration = len(audio_array) / original_sr
	if audio_duration > config["max_audio_sec"]:
	gr.Warning(f"Audio blocked! Exceeds the {config['max_audio_sec']} second limit.")
	return ""

	if audio_array.ndim > 1: audio_array = np.mean(audio_array, axis=1)
	audio_array = audio_array.astype(np.float32)
	if np.max(np.abs(audio_array)) > 1.0: audio_array = audio_array / 32768.0
	if original_sr != 16000: audio_array = librosa.resample(y=audio_array, orig_sr=original_sr, target_sr=16000)

	ai_lang = "english" if language_choice == "English" else "chinese"
	result = local_whisper({"sampling_rate": 16000, "raw": audio_array}, generate_kwargs={"language": ai_lang, "task": "transcribe"})
	return result['text']
	except Exception: return ""

	def agent_chat(user_text, chat_history):
	global knowledge_base
	try:
	if not user_text or not str(user_text).strip():
	gr.Warning("⚠️ 請輸入文字！")
	return "", chat_history

	config = load_config()
	user_input_str = str(user_text) # 強制轉成字串，防止意外

	# --- 1. 路由判斷 (確保格式正確) ---
	router_prompt = f"Determine if this is SCHOOL or GENERAL: {user_input_str}"
	route_res = client.chat_completion(
	model=config["llm_model"],
	messages=[{"role": "user", "content": router_prompt}], # 這裡必須是字典
	max_tokens=5
	)

	# --- 2. 取得 System Prompt ---
	route_verdict = route_res.choices[0].message.content.strip()
	if "SCHOOL" in route_verdict:
	if knowledge_base is not None:
	docs = knowledge_base.similarity_search(user_input_str, k=2)
	context = "\n".join([doc.page_content for doc in docs])
	system_prompt = f"{config['prompt_school']}\nContext: {context}"
	else:
	system_prompt = "You are a school assistant. No documents found."
	else:
	system_prompt = config['prompt_general']

	# --- 3. 關鍵修正：構建發送給 AI 的完整清單 ---
	# 我們在這裡明確定義 role 和 content，確保完全符合 API 要求
	final_messages = [
	{"role": "system", "content": str(system_prompt)}
	]

	# 如果你想加入記憶（History），請用這個迴圈：
	for old_user, old_ai in chat_history[-5:]: # 取最近5次對話
	final_messages.append({"role": "user", "content": str(old_user)})
	final_messages.append({"role": "assistant", "content": str(old_ai)})

	# 加入本次的問題
	final_messages.append({"role": "user", "content": user_input_str})

	# --- 4. 發送請求 ---
	final_response = client.chat_completion(
	model=config["llm_model"],
	messages=final_messages, # 傳入剛才建好的字典清單
	max_tokens=200,
	temperature=0.5
	)

	ai_reply = final_response.choices[0].message.content
	chat_history.append((user_input_str, ai_reply))
	return "", chat_history

	except Exception as e:
	print(f"DEBUG ERROR: {e}") # 在後台印出真正的錯誤原因
	chat_history.append(("Error", f"⚠️ 格式錯誤或連線中斷: {str(e)}"))
	return "", chat_history
	# ------------------------------------------
	# 3. ADMIN SETTINGS HELPERS
	# ------------------------------------------
	def admin_login(u, p):
	if u == "admin" and p == "admin": return gr.update(visible=False), gr.update(visible=True)
	return gr.update(visible=True), gr.update(visible=False)

	def update_ai_settings(model, p_school, p_general):
	conf = load_config()
	final_model = model.strip() if model and model.strip() != "" else "Qwen/Qwen2.5-7B-Instruct"
	conf.update({"llm_model": final_model, "prompt_school": p_school, "prompt_general": p_general})
	save_config(conf)
	gr.Info(f"✅ AI Config Saved! Model set to: {final_model}")
	return final_model

	def update_security_settings(audio_sec, file_mb):
	conf = load_config()
	conf.update({"max_audio_sec": audio_sec, "max_file_mb": file_mb})
	save_config(conf)
	gr.Info("🛡️ Security Guards updated successfully!")

	def export_project():
	zip_path = os.path.join(EXPORT_DIR, "school_bot_deploy.zip")
	req_text = "gradio\ntransformers\ntorch\nlibrosa\nsoundfile\ndocx2txt\nlangchain\nlangchain-community\nlangchain-huggingface\nlangchain-text-splitters\nfaiss-cpu\npypdf\nhuggingface_hub"
	with zipfile.ZipFile(zip_path, 'w') as zf:
	zf.writestr("requirements.txt", req_text)
	zf.writestr("README.md", "# Deployment Guide\n1. Upload this code to Hugging Face Spaces.\n2. Add your HF_TOKEN to the Space Secrets.\n3. Run app.py!")
	return gr.update(visible=True, value=zip_path)

	# ------------------------------------------
	# 4. THE UI LAYOUT
	# ------------------------------------------
	print("Building Final UI...")
	current_config = load_config()

	with gr.Blocks(theme=gr.themes.Soft()) as demo:

	with gr.Tab("💬 Student Chatbot"):
	gr.Markdown("# 🦸‍♂️ My Ultimate Hybrid AI Agent")
	chatbot = gr.Chatbot(height=400)
	with gr.Row():
	with gr.Column(scale=3): txt_input = gr.Textbox(label="Type a message")
	with gr.Column(scale=2):
	lang_toggle = gr.Radio(choices=["English", "Chinese"], value="Chinese", label="Voice Language:")
	aud_input = gr.Audio(sources=["microphone"], type="numpy", label="Record Voice")
	with gr.Row():
	submit_btn = gr.Button("🚀 Send to AI", variant="primary")
	clear_btn = gr.ClearButton([txt_input, aud_input, chatbot], value="🗑️ Clear Chat")

	aud_input.change(fn=transcribe_only, inputs=[aud_input, lang_toggle], outputs=[txt_input])
	submit_btn.click(fn=agent_chat, inputs=[txt_input, chatbot], outputs=[txt_input, chatbot])

	with gr.Tab("⚙️ Admin Settings"):
	with gr.Column(visible=True) as login_screen:
	gr.Markdown("### 🔒 Admin Login")
	with gr.Row():
	usr_box = gr.Textbox(label="Username")
	pwd_box = gr.Textbox(label="Password", type="password")
	login_btn = gr.Button("Login", variant="primary")

	with gr.Column(visible=False) as admin_panel:
	gr.Markdown("## 🎛️ Platform Control Center")

	with gr.Tabs():
	with gr.Tab("📂 Data Manager"):
	with gr.Row():
	# THE UI UPGRADE: Now accepts .docx, .csv, and .md!
	upload_btn = gr.File(
	label="1. Upload File (Click or Drag & Drop)",
	file_count="multiple",
	type="filepath",
	file_types=[".pdf", ".txt", ".docx", ".csv", ".md"]
	)
	with gr.Column():
	download_btn = gr.Button("2. ⬇️ Download Selected")
	delete_btn = gr.Button("3. 🗑️ Delete Selected", variant="stop")
	download_output = gr.File(label="Click to Save", visible=False)
	gr.Markdown("---")
	with gr.Row():
	gr.Markdown("### 📄 Database List View")
	select_all_btn = gr.Button("☑️ Select All", size="sm")
	file_listview = gr.CheckboxGroup(choices=get_file_list(), label="")

	with gr.Tab("🧠 AI Config & Prompts"):
	gr.Markdown("Change how the AI behaves in real-time without restarting the server.")
	ai_model_txt = gr.Textbox(value=current_config["llm_model"], label="Cloud LLM Model (Hugging Face ID)", placeholder="Leave blank to use default (Qwen/Qwen2.5-7B-Instruct)")
	prompt_school_txt = gr.TextArea(value=current_config["prompt_school"], label="School RAG System Prompt")
	prompt_general_txt = gr.TextArea(value=current_config["prompt_general"], label="General Conversation System Prompt")
	save_ai_btn = gr.Button("💾 Save AI Settings", variant="primary")

	with gr.Tab("🛡️ Security Guards"):
	gr.Markdown("Protect your server from spam and massive files.")
	audio_limit = gr.Slider(minimum=5, maximum=60, step=1, value=current_config["max_audio_sec"], label="Max Voice Recording Length (Seconds)")
	file_limit = gr.Slider(minimum=1, maximum=50, step=1, value=current_config["max_file_mb"], label="Max Upload File Size (MB)")
	save_sec_btn = gr.Button("💾 Save Security Settings", variant="primary")

	with gr.Tab("☁️ Cloud Deployment"):
	gr.Markdown("### Ready to go live 24/7?")
	gr.Markdown("Google Colab goes to sleep when you close your browser. To keep this bot online forever, export your deployment kit and upload it to a free host like Hugging Face Spaces.")
	export_btn = gr.Button("📦 Generate Deployment Kit (ZIP)", variant="primary")
	export_output = gr.File(label="Download your ZIP", visible=False)

	# WIRING
	login_btn.click(fn=admin_login, inputs=[usr_box, pwd_box], outputs=[login_screen, admin_panel])
	upload_btn.upload(fn=upload_files, inputs=[upload_btn], outputs=[file_listview, upload_btn])
	delete_btn.click(fn=delete_selected_files, inputs=[file_listview], outputs=[file_listview])
	download_btn.click(fn=download_selected_files, inputs=[file_listview], outputs=[download_output])
	def select_all(): return gr.update(value=get_file_list())
	select_all_btn.click(fn=select_all, inputs=[], outputs=[file_listview])
	save_ai_btn.click(fn=update_ai_settings, inputs=[ai_model_txt, prompt_school_txt, prompt_general_txt], outputs=[ai_model_txt])
	save_sec_btn.click(fn=update_security_settings, inputs=[audio_limit, file_limit])
	export_btn.click(fn=export_project, outputs=[export_output])

	demo.launch(share=True)