Updated app.py with recommendation

app.py

@@ -11,11 +11,20 @@ from contextlib import asynccontextmanager  # Add this import
 from models.rice_model import RiceDiseaseCNN
 from models.rice_leaf_validator import RiceLeafValidator, is_rice_leaf
 from utils.image_processing import process_image
+
 import requests
 import json
 import os
 from typing import Dict, Any
 
+# Hugging Face transformers pipeline for text recommendations
+try:
+    from transformers import pipeline
+    recommender = pipeline("text-generation", model="google/flan-t5-small")
+except Exception as e:
+    recommender = None
+    logging.warning(f"Transformers pipeline not loaded: {e}")
+
 # Configure logging
 logging.basicConfig(
     level=logging.INFO,
@@ -87,56 +96,16 @@ RECOMMENDATIONS = {
     }
 }
 
-def get_ollama_recommendation(disease_name: str, confidence: float) -> Dict[str, Any]:
-    """Get AI-generated recommendation using Ollama"""
-    try:
-        logging.info(f"Requesting Ollama recommendation for {disease_name} with {confidence:.2f}% confidence")
-
-        # Create a detailed prompt
-        prompt = f"""
-        As an agricultural expert specializing in rice diseases in the Philippines:
-
-        Our AI system has detected {disease_display_names.get(disease_name, disease_name)} with {confidence:.2f}% confidence in a rice plant image.
-
-        Please provide:
-        1. A detailed analysis of how this disease affects rice plants
-        2. Specific recommendations for treating this disease infection
-        3. Preventive measures farmers should take to avoid future infections
 
-        Format your response in clear sections that can be easily read by farmers.
-        """
-
-        # Call local Ollama API
-        response = requests.post('http://localhost:11434/api/generate',
-            json={
-                "model": "mistral:7b-instruct",
-                "prompt": prompt,
-                "stream": False
-            },
-            timeout=60 # 60-second timeout
-        )
-
-        if response.status_code == 200:
-            result = response.json()
-            ai_recommendation = result.get("response", "")
-
-            # Add this debugging
-            logging.info(f"Ollama response for {disease_name} (first 100 chars): {ai_recommendation[:100]}...")
-
-            return {
-                "recommendation": ai_recommendation,
-                "source": "ai",
-                "details": f"AI-generated recommendation for {disease_name}"
-            }
-        else:
-            logging.error(f"Ollama API error: {response.status_code}")
-            # Fall back to static recommendation
-            return RECOMMENDATIONS.get(disease_name, {})
-
-    except Exception as e:
-        logging.error(f"Error calling Ollama: {e}")
-        # Fall back to static recommendation
-        return RECOMMENDATIONS.get(disease_name, {})
+# New Hugging Face recommendation endpoint
+@app.post("/recommend/")
+async def recommend(disease: str):
+    """Generate farming recommendations for a rice disease using Hugging Face LLM"""
+    if recommender is None:
+        return {"error": "Text generation pipeline not available."}
+    prompt = f"Give 2-3 short farming recommendations for managing {disease} in rice plants."
+    response = recommender(prompt, max_length=100, num_return_sequences=1)
+    return {"recommendation": response[0]['generated_text']}
 
 def load_model():
     """Load the PyTorch models"""