import gradio as gr
import numpy as np, re, os, joblib
from sentence_transformers import SentenceTransformer
from huggingface_hub import hf_hub_download
import fitz, docx, chardet

MODEL_REPO = "kauzan25/pathora-bert-classifier"
print("Loading...")
model_path = hf_hub_download(repo_id=MODEL_REPO, filename="artifacts/sklearn_model.joblib", repo_type="model")
le_path = hf_hub_download(repo_id=MODEL_REPO, filename="artifacts/label_encoder.joblib", repo_type="model")
clf = joblib.load(model_path)
le = joblib.load(le_path)
st = SentenceTransformer("all-MiniLM-L6-v2")
print("Loaded!")

def process(file, text_input):
    if file is not None:
        ext = os.path.splitext(file.name)[1].lower()
        if ext == ".pdf":
            doc = fitz.open(file.name)
            text = "".join(p.get_text() for p in doc); doc.close()
        elif ext in (".doc", ".docx"):
            doc = docx.Document(file.name)
            text = chr(10).join(p.text for p in doc.paragraphs)
        elif ext == ".txt":
            with open(file.name, "rb") as f:
                raw = f.read()
            text = raw.decode(chardet.detect(raw)["encoding"] or "utf-8", errors="ignore")
        else:
            return "Unsupported.", None
        src = ext.upper().lstrip(".")
    elif text_input and text_input.strip():
        text = text_input.strip(); src = "Text"
    else:
        return "Upload file or enter text.", None
    if len(text) < 20:
        return "Text too short.", None
    text = re.sub("<[^>]+>", " ", text)
    text = re.sub("\\s+", " ", text).strip()
    emb = st.encode([text], convert_to_numpy=True)
    probs = clf.predict_proba(emb)[0]
    top = np.argsort(probs)[::-1][:5]
    results = [{"category": le.classes_[i], "confidence": float(probs[i])} for i in top]
    out = "### Hasil: **{}** ({:.1f}%)".format(results[0]["category"], results[0]["confidence"]*100)
    for i, p in enumerate(results):
        out += chr(10) + "{}. {} ({:.1f}%)".format(i+1, p["category"], p["confidence"]*100)
    return out, {"predictions": results}

with gr.Blocks(title="PathOra") as demo:
    gr.Markdown("# PathOra - Career Prediction")
    f_in = gr.File(label="File", file_types=[".pdf",".doc",".docx",".txt"])
    t_in = gr.Textbox(label="Or paste text", lines=3)
    btn = gr.Button("Analyze")
    out_md = gr.Markdown()
    out_json = gr.JSON()
    btn.click(process, [f_in, t_in], [out_md, out_json])

demo.launch()