Update rag.py

rag.py

@@ -6,6 +6,7 @@ import os
 import pandas as pd
 from datasets import load_dataset, Dataset
 from dotenv import load_dotenv
+import random
 import glob
 
 # Load environment variables
@@ -28,7 +29,15 @@ GREETINGS = [
     "hey there", "greetings"
 ]
 
-# Load multiple JSON datasets from the datasets folder
+# Fixed rephrased unmatched query responses
+UNMATCHED_RESPONSES = [
+    "Thank you for your query. We’ve forwarded it to our support team and it will be added soon. In the meantime, you can visit the University of Education official website or reach out via the contact details below.\n\n📞 +92-42-99262231-33\n✉️ info@ue.edu.pk\n🌐 https://ue.edu.pk",
+    "We’ve noted your question and it’s in queue for inclusion. For now, please check the University of Education website or contact the administration directly.\n\n📞 +92-42-99262231-33\n✉️ info@ue.edu.pk\n🌐 https://ue.edu.pk",
+    "Your query has been recorded. We’ll update the system with relevant information shortly. Meanwhile, you can visit UE's official site or reach out using the details below:\n\n📞 +92-42-99262231-33\n✉️ info@ue.edu.pk\n🌐 https://ue.edu.pk",
+    "We appreciate your question. It has been forwarded for further processing. Until it’s available here, feel free to visit the official UE website or use the contact options:\n\n📞 +92-42-99262231-33\n✉️ info@ue.edu.pk\n🌐 https://ue.edu.pk"
+]
+
+# Load multiple JSON datasets
 dataset = []
 try:
     json_files = glob.glob('datasets/*.json')
@@ -49,10 +58,7 @@ except Exception as e:
 # Precompute embeddings
 dataset_questions = [item.get("Question", "").lower().strip() for item in dataset]
 dataset_answers = [item.get("Answer", "") for item in dataset]
-if dataset_questions:
-    dataset_embeddings = similarity_model.encode(dataset_questions, convert_to_tensor=True)
-else:
-    dataset_embeddings = None
+dataset_embeddings = similarity_model.encode(dataset_questions, convert_to_tensor=True)
 
 # Save unmatched queries to Hugging Face
 def manage_unmatched_queries(query: str):
@@ -63,7 +69,6 @@ def manage_unmatched_queries(query: str):
             df = ds["train"].to_pandas()
         except:
             df = pd.DataFrame(columns=["Query", "Timestamp", "Processed"])
-        
         if query not in df["Query"].values:
             new_entry = {"Query": query, "Timestamp": timestamp, "Processed": False}
             df = pd.concat([df, pd.DataFrame([new_entry])], ignore_index=True)
@@ -89,18 +94,16 @@ def query_groq_llm(prompt, model_name="llama3-70b-8192"):
         print(f"Error querying Groq API: {e}")
         return ""
 
-# Main logic function
+# Main logic function to be called from Gradio
 def get_best_answer(user_input):
     if not user_input.strip():
         return "Please enter a valid question."
 
     user_input_lower = user_input.lower().strip()
 
-    # Basic length check unless it's a greeting
     if len(user_input_lower.split()) < 3 and not any(greet in user_input_lower for greet in GREETINGS):
         return "Please ask your question properly with at least 3 words."
 
-    # Specific Keyword Check for Fees
     if any(keyword in user_input_lower for keyword in ["fee structure", "fees structure", "semester fees", "semester fee"]):
         return (
             "💰 For complete and up-to-date fee details for this program, we recommend visiting the official University of Education fee structure page.\n"
@@ -108,53 +111,33 @@ def get_best_answer(user_input):
             "🔗 https://ue.edu.pk/allfeestructure.php"
         )
 
-    # Calculate Similarity
     user_embedding = similarity_model.encode(user_input_lower, convert_to_tensor=True)
     similarities = util.pytorch_cos_sim(user_embedding, dataset_embeddings)[0]
     best_match_idx = similarities.argmax().item()
     best_score = similarities[best_match_idx].item()
 
-    # LOGIC BRANCHING
-    if best_score >= 0.65:
-        # PATH 1: Dataset Match - Rephrase the Answer
-        original_answer = dataset_answers[best_match_idx]
-        prompt = f"""Name is UOE AI Assistant! You are an official assistant for the University of Education Lahore.
+    if best_score < 0.65:
+        manage_unmatched_queries(user_input)
+        return random.choice(UNMATCHED_RESPONSES)
+
+    original_answer = dataset_answers[best_match_idx]
+    prompt = f"""Name is UOE AI Assistant! You are an official assistant for the University of Education Lahore.
 Rephrase the following official answer clearly and professionally. 
-Use structured formatting (like headings, bullet points, or numbered lists) where appropriate to make it attractive.
+Use structured formatting (like headings, bullet points, or numbered lists) where appropriate.
 DO NOT add any new or extra information. ONLY rephrase and improve the clarity and formatting of the original answer.
 ### Question:
 {user_input}
 ### Original Answer:
 {original_answer}
 ### Rephrased Answer:
-"""
-    else:
-        # PATH 2: No Match - Answer from LLM Knowledge and Log
-        manage_unmatched_queries(user_input)
-        prompt = f"""Name is UOE AI Assistant! As an official assistant for University of Education Lahore, provide a helpful and professional response based on university standards.
-Include relevant details about university policies if known. 
-If unsure about specific dates or numbers, direct the user to official channels.
-### Question:
-{user_input}
-### Official Answer:
 """
 
     llm_response = query_groq_llm(prompt)
 
     if llm_response:
-        # Clean up markers if the LLM includes them in the output
         for marker in ["Improved Answer:", "Official Answer:", "Rephrased Answer:"]:
             if marker in llm_response:
                 return llm_response.split(marker)[-1].strip()
         return llm_response
     else:
-        # Fallback if Groq fails
-        if best_score >= 0.65:
-            return dataset_answers[best_match_idx]
-        else:
-            return (
-                "For official information:\n"
-                "📞 +92-42-99262231-33\n"
-                "✉️ info@ue.edu.pk\n"
-                "🌐 https://ue.edu.pk"
-            )
+        return dataset_answers[best_match_idx]