import os
import re
import traceback
import numpy as np
import gradio as gr
from openai import OpenAI

# ======== 1. HF Secret 讀取與安全檢查 ========
secret_key = os.environ.get("OPENAIAPIKEY")
if not secret_key:
    raise ValueError(
        "⚠️ HF Space Secret 'OPENAIAPIKEY' 未設定或名稱錯誤！"
        "請先在 HF Space Secrets 裡建立此 Secret 並重新部署。"
    )
os.environ["OPENAI_API_KEY"] = secret_key

# ======== 2. 初始化 OpenAI SDK ========
client = OpenAI()

# ======== 3. 專業領域定義 ========
PROFESSIONS = {
    "程式設計": "你是一位資深程式設計師，回答必須專業、詳細，附上程式範例與步驟。",
    "行銷": "你是一位行銷專家，回答必須專業、詳細，提供可執行行銷策略與步驟。",
    "法律": "你是一位律師，回答必須專業、法律依據明確、可操作建議清楚。",
    "醫療": "你是一位醫師，回答必須專業、以健康與安全為前提，提供可操作建議。",
    "財務": "你是一位財務顧問，回答必須專業、符合台灣會計與稅務法規，提供可執行建議。",
    "設計": "你是一位設計師，回答必須專業、詳細，提供設計步驟與案例。"
}

# ======== 4. UTF-8 安全函數 ========
def safe_utf8(text):
    if not isinstance(text, str):
        text = str(text)
    return text.encode("utf-8", errors="ignore").decode("utf-8")

# ======== 5. Embedding 初始化 ========
profession_embeddings = {}

def ensure_embeddings():
    global profession_embeddings
    if not profession_embeddings:
        for field in PROFESSIONS.keys():
            try:
                emb = client.embeddings.create(
                    model="text-embedding-3-small",
                    input=safe_utf8(field)
                )
                profession_embeddings[field] = np.array(emb.data[0].embedding)
            except Exception as e:
                print(f"Embedding 初始化失敗 ({field}):", e)
                profession_embeddings[field] = np.zeros(1536)

# ======== 6. 自動判斷職業 ========
def detect_profession(user_input: str) -> str:
    ensure_embeddings()
    try:
        text_emb = client.embeddings.create(
            model="text-embedding-3-small",
            input=safe_utf8(user_input)
        )
        text_vector = np.array(text_emb.data[0].embedding)
    except Exception as e:
        print("Embedding 呼叫錯誤:", traceback.format_exc())
        return "你是一個專業顧問，回答必須專業、詳細、可操作。"

    scores = {}
    for field, emb in profession_embeddings.items():
        if np.linalg.norm(text_vector) == 0 or np.linalg.norm(emb) == 0:
            scores[field] = -1
        else:
            scores[field] = np.dot(text_vector, emb) / (np.linalg.norm(text_vector)*np.linalg.norm(emb))
    best_field = max(scores, key=scores.get)
    return PROFESSIONS[best_field]

# ======== 7. AI Agent 回答 ========
def professional_agent(user_input, state):
    user_input = safe_utf8(user_input)

    if state.get("profession_prompt") is None:
        profession_prompt = detect_profession(user_input)
        state["profession_prompt"] = profession_prompt
        question = re.sub(r"我是.*?(，|,)", "", user_input)
        if not question.strip():
            answer = f"✅ 已設定你的專業領域：\n{profession_prompt}\n請提出問題。"
            state["chat_history"].append({"role":"user","content":user_input})
            state["chat_history"].append({"role":"assistant","content":answer})
            return state["chat_history"], state
        else:
            user_input = question

    messages = [{"role": "system", "content": state["profession_prompt"]}]
    for msg in state["chat_history"]:
        messages.append({"role": msg["role"], "content": msg["content"]})
    messages.append({"role": "user", "content": user_input})

    try:
        response = client.chat.completions.create(
            model="gpt-4o-mini",
            messages=messages,
            temperature=0.2
        )
        answer = safe_utf8(response.choices[0].message.content)
    except Exception as e:
        tb = traceback.format_exc()
        print(tb)  # 後端完整 log
        answer = f"⚠️ 發生錯誤，請查看後端 logs: {str(e)}"

    # Append using dict 格式
    state["chat_history"].append({"role":"user","content":user_input})
    state["chat_history"].append({"role":"assistant","content":answer})

    # 保留最近 10 條訊息
    if len(state["chat_history"]) > 20:
        state["chat_history"] = state["chat_history"][-20:]

    return state["chat_history"], state

# ======== 8. Gradio 介面 ========
with gr.Blocks() as demo:
    gr.Markdown("## 🧑‍💼AI 專業領域顧問")
    gr.Markdown("第一次輸入可以同時輸入職業 + 問題，例如：我是會計師，我想知道台灣稅務分析")

    chatbot = gr.Chatbot(type="messages")
    msg = gr.Textbox(label="輸入訊息")
    state = gr.State({"chat_history": [], "profession_prompt": None})

    msg.submit(professional_agent, [msg, state], [chatbot, state])

demo.launch(server_name="0.0.0.0", server_port=7860)