Spaces:
Sleeping
Sleeping
Download main.py from nextstepv1/nextstep-ai-api: direct link, hf CLI and curl.
- Browser
- Download file 9.41 kB
-
https://huggingface.co/spaces/nextstepv1/nextstep-ai-api/resolve/main/main.py
- Command line
-
hf download hf://spaces/nextstepv1/nextstep-ai-api/main.py
-
curl -L -o main.py https://huggingface.co/spaces/nextstepv1/nextstep-ai-api/resolve/main/main.py
9.41 kB
| from fastapi import FastAPI, UploadFile, File | |
| from deep_translator import GoogleTranslator | |
| import pandas as pd | |
| import numpy as np | |
| import pickle | |
| import tensorflow as tf | |
| import PyPDF2 | |
| import io | |
| import urllib.parse | |
| import os | |
| from groq import Groq | |
| original_dense_init = tf.keras.layers.Dense.__init__ | |
| def safe_dense_init(self, *args, **kwargs): | |
| kwargs.pop('quantization_config', None) | |
| original_dense_init(self, *args, **kwargs) | |
| tf.keras.layers.Dense.__init__ = safe_dense_init | |
| app = FastAPI(title="NextStep AI API") | |
| async def root(): | |
| return {"message": "Selamat datang di NextStep AI API! Silakan akses /docs untuk dokumentasi."} | |
| print(" Memuat Dataset, Model & AI...") | |
| groq_api_key = os.getenv("GROQ_API_KEY") | |
| if groq_api_key: | |
| groq_client = Groq(api_key=groq_api_key) | |
| else: | |
| groq_client = None | |
| print("Peringatan: GROQ_API_KEY tidak ditemukan") | |
| # LOAD DATASET | |
| df_job = pd.read_csv('Dataset/english_jobs.csv') | |
| print("-> Kolom di english_jobs.csv:", df_job.columns.tolist()) | |
| try: | |
| if os.path.exists('Dataset/jobs.csv'): | |
| df_metadata = pd.read_csv('Dataset/jobs.csv') | |
| df_metadata_clean = df_metadata.drop_duplicates(subset=['title'])[['title', 'company', 'url']] | |
| df_job = pd.merge( | |
| left=df_job, | |
| right=df_metadata_clean, | |
| left_on='raw_job_title', | |
| right_on='title', | |
| how='left' | |
| ) | |
| print("-> Sukses! Berhasil menggabungkan dataset dengan jobs.csv!") | |
| except Exception as e: | |
| print(f"-> Peringatan: Gagal menggabungkan jobs.csv: {e}") | |
| df_job = df_job.dropna(subset=['extracted_skills', 'job_category']) | |
| # LOAD MODEL KLASIFIKASI | |
| with open('models/tfidf_text.pkl', 'rb') as f: | |
| tfidf_text = pickle.load(f) | |
| with open('models/mlb.pkl', 'rb') as f: | |
| mlb = pickle.load(f) | |
| model_klasifikasi = tf.keras.models.load_model('models/best_model.h5') | |
| with open('models/model_prediksi_kategori.pkl', 'rb') as f: | |
| model_kategori = pickle.load(f) | |
| # LOAD MODEL SIAMESE | |
| class CosineSimilarityLayer(tf.keras.layers.Layer): | |
| def __init__(self, **kwargs): | |
| super(CosineSimilarityLayer, self).__init__(**kwargs) | |
| def call(self, inputs): | |
| cv_vec, job_vec = inputs | |
| cv_norm = tf.nn.l2_normalize(cv_vec, axis=-1) | |
| job_norm = tf.nn.l2_normalize(job_vec, axis=-1) | |
| return tf.reduce_sum(cv_norm * job_norm, axis=-1, keepdims=True) | |
| with open('Siamese/vectorizer_siamese_final.pkl', 'rb') as f: | |
| vectorizer_siamese = pickle.load(f) | |
| with open('Siamese/pca_siamese_final.pkl', 'rb') as f: | |
| pca_siamese = pickle.load(f) | |
| recommender_model = tf.keras.models.load_model( | |
| 'Siamese/recommender_model_final.keras', | |
| custom_objects={'CosineSimilarityLayer': CosineSimilarityLayer} | |
| ) | |
| print("Semua Model Berhasil Dimuat!") | |
| # HELPER FUNCTIONS | |
| def ekstrak_teks_pdf(file_bytes): | |
| teks_lengkap = "" | |
| reader = PyPDF2.PdfReader(io.BytesIO(file_bytes)) | |
| for page in reader.pages: | |
| teks_lengkap += page.extract_text() + " " | |
| return teks_lengkap | |
| def terjemahkan_cv(teks): | |
| try: | |
| if len(teks) > 4900: | |
| teks = teks[:4900] | |
| return GoogleTranslator(source='auto', target='en').translate(teks) | |
| except Exception as e: | |
| print(f"Translasi Gagal (Internet down/limit): {e}") | |
| return teks | |
| def cari_skill_gap(teks_cv, syarat_skill): | |
| clean_syarat = str(syarat_skill).replace('[','').replace(']','').replace("'", "").split(',') | |
| gap = [skill.strip() for skill in clean_syarat if skill.strip().lower() not in teks_cv.lower()] | |
| return [g for g in gap if g] | |
| # ENDPOINT API | |
| async def predict_cv(file: UploadFile = File(...)): | |
| try: | |
| file_bytes = await file.read() | |
| teks_cv = ekstrak_teks_pdf(file_bytes) | |
| vektor_cv_ai = tfidf_text.transform([teks_cv]).toarray() | |
| vektor_dummy = np.zeros((1, 50)) | |
| vektor_input_klasifikasi = [vektor_cv_ai, vektor_dummy] | |
| raw_pred = model_klasifikasi.predict(vektor_input_klasifikasi, verbose=0) | |
| if isinstance(raw_pred, list): | |
| raw_pred = raw_pred[0] | |
| prediksi_prob = np.array(raw_pred).flatten() | |
| y_pred_binary = (prediksi_prob > 0.50).astype(int) | |
| skills_pelamar = mlb.inverse_transform(y_pred_binary.reshape(1, -1))[0] | |
| list_skill_pelamar = list(skills_pelamar) | |
| kategori_prediksi = model_kategori.predict(vektor_cv_ai)[0] | |
| df_filtered = df_job[df_job['job_category'] == kategori_prediksi] | |
| if df_filtered.empty: | |
| df_filtered = df_job | |
| df_sample = df_filtered.sample(min(100, len(df_filtered))) | |
| array_mentah = vectorizer_siamese.transform([teks_cv]).toarray() | |
| vektor_input_siamese = pca_siamese.transform(array_mentah) | |
| hasil_rekomendasi = [] | |
| for index, row in df_sample.iterrows(): | |
| teks_job = str(row['extracted_skills']) | |
| array_job = vectorizer_siamese.transform([teks_job]).toarray() | |
| vektor_job = pca_siamese.transform(array_job) | |
| skor_ai = recommender_model.predict([vektor_input_siamese, vektor_job], verbose=0)[0][0] | |
| persentase = round(float(((skor_ai + 1) / 2) * 100), 2) | |
| posisi_loker = row.get('raw_job_title', f"Lowongan {index}") | |
| perusahaan_loker = row.get('company') | |
| if pd.isna(perusahaan_loker) or str(perusahaan_loker).strip() == "": | |
| perusahaan_loker = "Perusahaan Mitra" | |
| link_asli = row.get('url') | |
| if pd.isna(link_asli) or str(link_asli).strip() == "": | |
| if perusahaan_loker == "Perusahaan Mitra": | |
| kata_kunci = str(posisi_loker) | |
| else: | |
| kata_kunci = f"{posisi_loker} {perusahaan_loker}" | |
| link_lamaran = f"https://www.linkedin.com/jobs/search/?keywords={urllib.parse.quote(kata_kunci)}" | |
| else: | |
| link_lamaran = str(link_asli) | |
| hasil_rekomendasi.append({ | |
| "posisi": posisi_loker if pd.notna(posisi_loker) else f"Lowongan {index}", | |
| "perusahaan": perusahaan_loker, | |
| "industri": kategori_prediksi, | |
| "skor_match": persentase, | |
| "syarat_skill": teks_job, | |
| "link_lamaran": link_lamaran | |
| }) | |
| hasil_rekomendasi = sorted(hasil_rekomendasi, key=lambda x: x['skor_match'], reverse=True)[:3] | |
| final_rekomendasi = [] | |
| for loker in hasil_rekomendasi: | |
| gap = cari_skill_gap(teks_cv, loker['syarat_skill']) | |
| loker['saran_belajar'] = gap[:5] | |
| #PROSES GENERASI REVIEW DARI GROQ | |
| try: | |
| prompt = f"""Kamu adalah Senior Recruiter dan Praktisi Ahli di industri {loker['industri']}. | |
| Keahlian pelamar saat ini: {list_skill_pelamar} | |
| Posisi yang dilamar: {loker['posisi']} di {loker['perusahaan']} | |
| Kekurangan skill (Gap): {gap[:5]} | |
| Tugas: Buat maksimal 3 kalimat evaluasi profesional berbahasa Indonesia. Analisis bagaimana keahlian pelamar saat ini dapat diintegrasikan dengan skill gap tersebut. Jelaskan fungsi spesifik atau peran dari skill gap tersebut dalam standar operasional (workflow) nyata untuk posisi {loker['posisi']}. | |
| Aturan: Gunakan terminologi teknis atau istilah profesional yang sangat akurat sesuai dengan industri {loker['industri']}. Langsung berikan evaluasi analitis tanpa basa-basi atau salam pembuka.""" | |
| chat_completion = groq_client.chat.completions.create( | |
| messages=[{"role": "user", "content": prompt}], | |
| model="llama-3.1-8b-instant", | |
| temperature=0.7, | |
| max_tokens=150 | |
| ) | |
| review = chat_completion.choices[0].message.content.strip() | |
| loker['review_deskriptif'] = review | |
| except Exception as e: | |
| print(f"Peringatan: Gagal mendapatkan review dari Groq: {e}") | |
| loker['review_deskriptif'] = f"Kamu memiliki dasar yang baik. Untuk memaksimalkan peluangmu sebagai {loker['posisi']}, sangat disarankan untuk mempelajari {', '.join(gap[:3])}." | |
| # Membuat link Coursera | |
| link_belajar = [] | |
| for skill in gap[:5]: | |
| url_belajar = f"https://www.coursera.org/search?query={urllib.parse.quote(skill)}" | |
| link_belajar.append(url_belajar) | |
| loker['link_belajar'] = link_belajar | |
| del loker['syarat_skill'] | |
| final_rekomendasi.append(loker) | |
| return { | |
| "status": "success", | |
| "data_pelamar": { | |
| "skill_terdeteksi": list_skill_pelamar | |
| }, | |
| "rekomendasi_utama_bidang": kategori_prediksi, | |
| "top_3_loker": final_rekomendasi | |
| } | |
| except Exception as e: | |
| return {"status": "error", "message": f"Terjadi kesalahan pada server: {str(e)}"} |