Backend

app.py

@@ -0,0 +1,88 @@
+import re
+import gradio as gr
+from transformers import pipeline
+
+# Load Hugging Face zero-shot classifier
+MODEL = "valhalla/distilbart-mnli-12-1"
+classifier = pipeline("zero-shot-classification", model=MODEL)
+
+LABELS = ["urgent", "fear", "authority", "financial scam", "safe"]
+
+# Regex cues
+CUES = {
+    "urgency": [r"\burgent\b", r"\bverify now\b", r"\blast chance\b"],
+    "fear": [r"\bsuspended\b", r"\block(ed)?\b"],
+    "authority": [r"\bCEO\b", r"\badmin(istrator)?\b"],
+    "financial": [r"\bprize\b", r"\blottery\b", r"\bmoney\b"],
+}
+
+SAFE_PHRASES = [
+    "mandatory email service announcement",
+    "privacy policy",
+    "unsubscribe from these emails"
+]
+
+URL_PATTERN_GLOBAL = re.compile(r"(https?://[^\s]+)")
+
+def regex_analysis(text: str):
+    score, findings = 0, []
+    for cat, pats in CUES.items():
+        for p in pats:
+            matches = re.findall(p, text, re.I)
+            if matches:
+                findings.append(f"{cat} cue → {matches[0]}")
+                score += 20
+    return score, findings
+
+def hf_analysis(text: str):
+    res = classifier(text, LABELS)
+    top = list(zip(res["labels"], res["scores"]))
+    top.sort(key=lambda x: x[1], reverse=True)
+    findings, score = [], 0
+    for lbl, sc in top[:2]:
+        if lbl != "safe" and sc > 0.3:
+            findings.append(f"HuggingFace: {lbl} ({sc:.2f})")
+            score += int(sc * 30)
+    return score, findings
+
+def extract_urls(text: str):
+    return list(set(URL_PATTERN_GLOBAL.findall(text)))
+
+def analyze_text(text: str):
+    regex_score, regex_findings = regex_analysis(text)
+    hf_score, hf_findings = hf_analysis(text)
+    urls = extract_urls(text)
+
+    score = min(100, regex_score + hf_score)
+    reasons = regex_findings + hf_findings
+
+    for phrase in SAFE_PHRASES:
+        if phrase in text.lower():
+            reasons.append(f"Safe phrase: {phrase}")
+            score = max(0, score - 15)
+
+    risk = "Low"
+    if score >= 70:
+        risk = "High"
+    elif score >= 35:
+        risk = "Medium"
+
+    return f"""
+📊 Score: {score}/100
+⚠️ Risk: {risk}
+🔎 Reasons:
+- {chr(10).join(reasons) if reasons else "None"}
+🌐 URLs: {', '.join(urls) if urls else "None"}
+"""
+
+# Gradio UI
+demo = gr.Interface(
+    fn=analyze_text,
+    inputs=gr.Textbox(lines=12, placeholder="Paste suspicious email or message here..."),
+    outputs="text",
+    title="PhishGuard 🛡️",
+    description="Detects phishing risks using regex cues + Hugging Face zero-shot classification."
+)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,3 @@
+gradio
+transformers
+torch