Update app.py

app.py

@@ -5,20 +5,25 @@ from prometheus_client import Counter, generate_latest, CONTENT_TYPE_LATEST
 from transformers import pipeline
 import gradio as gr
 
+# ✅ Expose FastAPI app at top level
 app = FastAPI(title="Incident Triage Bot")
 
+# 📊 Prometheus metrics
 triage_requests = Counter("triage_requests_total", "Total triage requests")
 triage_errors = Counter("triage_errors_total", "Total triage errors")
 
+# 🧠 Load lightweight model (CPU only)
 triage_pipeline = pipeline(
     "text2text-generation",
     model="google/flan-t5-small",
-    device=-1  # CPU-only
+    device=-1  # Force CPU
 )
 
+# 📥 Request model
 class Incident(BaseModel):
     description: str
 
+# 🔍 Triage API
 @app.post("/triage")
 async def triage(incident: Incident):
     triage_requests.inc()
@@ -30,14 +35,17 @@ async def triage(incident: Incident):
         triage_errors.inc()
         return {"error": str(e)}
 
+# 📈 Metrics endpoint
 @app.get("/metrics")
 def metrics():
     return Response(generate_latest(), media_type=CONTENT_TYPE_LATEST)
 
+# 🏠 Root
 @app.get("/")
 def home():
     return {"message": "LLM Incident Triage Bot is running!"}
 
+# 🎨 Gradio UI
 def gradio_triage(description):
     try:
         triage_requests.inc()