# app.py  ← เปลี่ยนชื่อไฟล์เป็น app.py
# 1) force sane thread env BEFORE any other imports
import os
for k in ["OMP_NUM_THREADS", "OPENBLAS_NUM_THREADS", "MKL_NUM_THREADS", "NUMEXPR_MAX_THREADS"]:
    v = os.environ.get(k, "")
    try:
        if int(v) <= 0:
            os.environ[k] = "1"
    except Exception:
        os.environ[k] = "1"
os.environ["TOKENIZERS_PARALLELISM"] = "false"

# 2) now import the rest
import gradio as gr
from fastapi import FastAPI
from pydantic import BaseModel

from agent_pdfimages import VS              # VS ต้องเป็น lazy-load ตามที่แก้ไว้
from agent_ml import build_agent_multimodal as build_agent
from langchain_openai import ChatOpenAI
# (ตัดของที่ไม่ใช้เพื่อลดโหลด)
# from langchain.agents import AgentExecutor
# from langchain.memory import ConversationBufferMemory

OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")

# FastAPI (ไม่บังคับ แต่คงไว้ได้)
app = FastAPI(title="Breast Cancer RAG Assistant")
agent = build_agent()
llm_direct = ChatOpenAI(model="gpt-4o", temperature=0.2, openai_api_key=OPENAI_API_KEY)

class Ask(BaseModel):
    query: str

@app.post("/ask")
def ask(q: Ask):
    try:
        return {"answer": agent.run(q.query)}
    except Exception:
        return {"answer": direct_rag_answer(q.query)}

def direct_rag_answer(question: str) -> str:
    docs = VS.similarity_search(question, k=3)
    context = "\n\n".join([(d.page_content or "")[:800] for d in docs])
    srcs = []
    for d in docs:
        m = d.metadata or {}
        tag = m.get("source_file", "unknown")
        if m.get("year"):
            tag += f" ({m.get('year')})"
        if m.get("page") is not None:
            tag += f", p.{m.get('page')}"
        srcs.append(tag)
    prompt = (
        "You are an assistant specializing in imaging and molecular profiles of breast cancer. "
        "Answer based on the provided text chunks. If info is incomplete, say so.\n\n"
        f"[Retrieved Chunks]\n{context}\n\n[Question]\n{question}\n\n"
        "Answer in bullet points. Sources: " + "；".join(srcs)
    )
    return llm_direct.invoke(prompt).content

def chat_fn(message, history):
    try:
        return agent.run(message)
    except Exception:
        return direct_rag_answer(message)

# IMPORTANT for Spaces (Gradio SDK): expose `demo` at top-level; DO NOT launch
demo = gr.ChatInterface(
    fn=chat_fn,
    title="🔬 Breast Cancer Knowledge RAG Assistant",
    description="Retrieves evidence from local PDFs (and images if indexed) and answers with citations."
)
# ⛔️ ไม่มี demo.launch(...)