main.py

import io
import re
import os
import torch
import PyPDF2
from fastapi import FastAPI, UploadFile, File, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from sentence_transformers import SentenceTransformer, util

app = FastAPI()

# ==============================
# CORS (Allow semua untuk testing)
# ==============================
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

# ==============================
# LOAD MODEL
# ==============================

REPO_ID = "lilcoderi/cv-matcher-model"

try:
    model = SentenceTransformer(REPO_ID)
    model.eval()
except Exception as e:
    raise RuntimeError(f"Gagal load model: {str(e)}")

THRESHOLD = 0.59

# ==============================
# REGEX OPTIMIZED
# ==============================
RE_CLEAN = re.compile(r'[•\-*●▪◦☑]')
RE_SPACES = re.compile(r'\s+')
RE_NON_ALPHA = re.compile(r'[^\w\s]')

# ==============================
# TEXT PREPROCESSING
# ==============================

def clean_text(text: str) -> str:
    text = text.lower()
    text = RE_CLEAN.sub(' ', text)
    text = text.encode("ascii", "ignore").decode()
    text = RE_NON_ALPHA.sub(' ', text)
    return RE_SPACES.sub(' ', text).strip()


def standardize_education(text: str) -> str:
    edu_map = {
        r'\b(sarjana|s1|strata 1|universitas|politeknik|institut)\b': 's1',
        r'\b(diploma 3|d3|ahli madya)\b': 'd3',
        r'\b(sma|smk|stm|smu|ma|sekolah menengah)\b': 'sma_smk',
    }
    for pattern, replacement in edu_map.items():
        text = re.sub(pattern, replacement, text)
    return text


def clean_job_description(text: str) -> str:
    noise_patterns = [
        r'we are hiring',
        r'send us your cv',
        r'kirim cv anda',
        r'subjek:.*',
        r'lowongan ini dibuka sampai.*',
        r'format pdf'
    ]
    for pattern in noise_patterns:
        text = re.sub(pattern, '', text, flags=re.IGNORECASE)
    return text


# ==============================
# PDF READER
# ==============================

def extract_text_from_pdf(file_bytes, max_pages=3):
    try:
        pdf_reader = PyPDF2.PdfReader(io.BytesIO(file_bytes))
        text = ""
        pages_to_read = min(len(pdf_reader.pages), max_pages)

        for i in range(pages_to_read):
            content = pdf_reader.pages[i].extract_text()
            if content:
                text += content + " "

        return text.strip()

    except Exception:
        raise HTTPException(status_code=400, detail="Gagal membaca file PDF")


# ==============================
# HEALTH CHECK (penting buat HF)
# ==============================

@app.get("/")
def root():
    return {"status": "CV Matcher API Running"}


# ==============================
# MAIN ENDPOINT
# ==============================

@app.post("/match")
async def match_cvs(
    job_file: UploadFile = File(...),
    cv_files: list[UploadFile] = File(...)
):

    # ---------- JOB DESCRIPTION ----------
    job_raw = extract_text_from_pdf(await job_file.read(), max_pages=5)
    job_cleaned = clean_job_description(job_raw)
    job_final = standardize_education(clean_text(job_cleaned))

    if not job_final:
        raise HTTPException(status_code=400, detail="Job description kosong")

    # ---------- CV PROCESS ----------
    cv_texts_processed = []
    filenames = []

    for cv in cv_files:
        content = await cv.read()
        raw_text = extract_text_from_pdf(content, max_pages=3)
        processed_text = standardize_education(clean_text(raw_text))

        if processed_text:
            cv_texts_processed.append(processed_text)
            filenames.append(cv.filename)

    if not cv_texts_processed:
        raise HTTPException(status_code=400, detail="Tidak ada CV yang valid")

    # ---------- EMBEDDING ----------
    with torch.no_grad():
        job_embedding = model.encode(
            job_final,
            convert_to_tensor=True,
            normalize_embeddings=True
        )

        cv_embeddings = model.encode(
            cv_texts_processed,
            convert_to_tensor=True,
            normalize_embeddings=True
        )

        scores = util.cos_sim(job_embedding, cv_embeddings)[0]

    # ---------- RESULT ----------
    results = []

    for i in range(len(filenames)):
        score_val = float(scores[i])

        results.append({
            "filename": filenames[i],
            "score": round(score_val, 4),
            "percentage": round(score_val * 100, 2),
            "status": "Cocok" if score_val >= THRESHOLD else "Tidak Cocok"
        })

    results.sort(key=lambda x: x['score'], reverse=True)

    return {"results": results}