Update app.py

app.py

@@ -28,14 +28,27 @@ def predict():
             input_text = f"{item['category']} - {item['subcategory']} in {item['area']}. {item.get('comments', '')}"
 
             # Smart keyword override for critical incidents
-            text_lower = input_text.lower()
-            critical_keywords = [
-                "umuusok", "sunog", "amoy sunog", "spark", "kuryente",
-                "leak", "baha", "gas", "short circuit", "smoke"
-            ]
-            if any(word in text_lower for word in critical_keywords):
-                results.append({"priority_score": 5})
-                continue
+          text_lower = input_text.lower()
+
+# Critical and low-urgency keyword lists
+critical_keywords = [
+    "umuusok", "sunog", "amoy sunog", "spark", "kuryente",
+    "leak", "baha", "gas", "short circuit", "smoke"
+]
+
+low_keywords = [
+    "lightbulb", "bumbilya", "ilaw", "palitan", "replace bulb",
+    "painting", "door knob", "hinge", "minor", "cosmetic"
+]
+
+# ✅ Priority rules
+if any(word in text_lower for word in critical_keywords):
+    results.append({"priority_score": 5})
+    continue
+elif any(word in text_lower for word in low_keywords):
+    results.append({"priority_score": 2})
+    continue
+
 
             # Tokenize & predict
             inputs = tokenizer(input_text, return_tensors="pt", truncation=True, padding=True)
@@ -47,6 +60,12 @@ def predict():
             # ✅ Convert back to 1–5 scale
             priority_score = predicted_class + 1
 
+# Soft correction: cap too-high or too-low predictions
+if priority_score >= 5 and not any(word in text_lower for word in critical_keywords):
+    priority_score = 4  # reduce overly high unless it's critical
+elif priority_score <= 1:
+    priority_score = 2  # lift up if too low
+
             # ✅ Here's where this line goes
             results.append({"priority_score": priority_score})