Update app.py

app.py

@@ -1,34 +1,62 @@
 import streamlit as st
 from transformers import pipeline
+from PIL import Image
+import pytesseract
+from pdf2image import convert_from_bytes
 
-# App title
-st.set_page_config(page_title="Fake Job Detector", page_icon="🕵️‍♂️")
-st.title("Fake Job / Lie Detector")
-st.markdown("Enter a job description and check if it seems suspicious!")
-
-# Load small NLI model for zero-shot classification (CPU-friendly)
+# -------------------
+# Load Hugging Face text classifier
+# -------------------
 @st.cache_resource
-def load_model():
-    return pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
+def load_classifier():
+    classifier = pipeline("text-classification", model="distilbert-base-uncased-finetuned-sst-2-english")
+    return classifier
+
+classifier = load_classifier()
+
+# -------------------
+# App UI
+# -------------------
+st.title("Fake Job Detector")
+st.write("Paste job text or upload a screenshot/PDF to check if the job posting is Legitimate or Fake.")
+
+# Job text input
+job_text = st.text_area("Paste job description here:")
 
-classifier = load_model()
+# File upload (image or PDF)
+uploaded_file = st.file_uploader("Or upload a screenshot / PDF", type=["png","jpg","jpeg","pdf"])
 
-# Input from user
-job_description = st.text_area("Enter the job description:")
+extracted_text = ""
 
-# Label options
-labels = ["Legitimate", "Suspicious", "Fake", "Scam"]
+# Extract text from file if uploaded
+if uploaded_file:
+    if uploaded_file.type == "application/pdf":
+        images = convert_from_bytes(uploaded_file.read())
+        for img in images:
+            extracted_text += pytesseract.image_to_string(img) + "\n"
+    else:
+        img = Image.open(uploaded_file)
+        extracted_text = pytesseract.image_to_string(img)
+
+# Combine pasted text + extracted text
+full_text = job_text + "\n" + extracted_text
 
-# Button to check
-if st.button("Check Job"):
-    if not job_description.strip():
-        st.warning("Please enter a job description!")
+if st.button("Detect"):
+    if full_text.strip() == "":
+        st.warning("Please paste job text or upload a file!")
     else:
-        # Run zero-shot classification
-        results = classifier(job_description, candidate_labels=labels)
-        st.subheader("Prediction:")
-        # Show top label and scores
-        top_label = results["labels"][0]
-        score = results["scores"][0]
-        st.success(f"Most likely: **{top_label}** ({score*100:.2f}%)")
-        st.write("Full results:", results)
+        result = classifier(full_text)
+        label = result[0]['label']
+        score = result[0]['score']
+
+        # Map labels to Fake/Suspicious/Legit (simple demo)
+        # Here using SST-2 labels for demo; in real app, fine-tune model
+        if label == "NEGATIVE":
+            final_label = "Suspicious / Fake"
+        else:
+            final_label = "Legitimate"
+
+        st.success(f"Prediction: **{final_label}**")
+        st.info(f"Confidence: {score:.2f}")
+        st.subheader("Extracted Job Text:")
+        st.text_area("Text Extracted", full_text, height=200)