Update chatbot.py

chatbot.py

@@ -7,6 +7,7 @@ from dotenv import load_dotenv
 from messages import krishna_blessings, ayush_teasing, keyword_groups
 from ayush_messages import ayush_surprises
 from sentence_transformers import SentenceTransformer, util
+import joblib
 import numpy as np
 
 # Configure logging
@@ -20,8 +21,14 @@ if not HUGGINGFACE_API_TOKEN:
     logger.error("HUGGINGFACE_API_TOKEN not found in environment variables.")
     raise ValueError("HUGGINGFACE_API_TOKEN is required.")
 
-# Initialize sentence transformer model
-semantic_model = SentenceTransformer('all-MiniLM-L6-v2')
+# Lazy load sentence transformer model and embeddings
+semantic_model = None
+keyword_embeddings_cache = joblib.load('embeddings_cache.joblib')
+
+def init_semantic_model():
+    global semantic_model
+    if semantic_model is None:
+        semantic_model = SentenceTransformer('all-MiniLM-L6-v2')
 
 # AI model for fallback responses
 AI_MODELS = [
@@ -78,13 +85,13 @@ def analyze_sentiment(user_input):
             if isinstance(result, list) and result:
                 emotions = result[0]
                 top_emotion = max(emotions, key=lambda x: x["score"])["label"]
-                return top_emotion  # e.g., "joy", "sadness", "anger"
+                return top_emotion
         return "neutral"
     except Exception as e:
         logger.error(f"Error in analyze_sentiment: {str(e)}")
         return "neutral"
 
-def make_api_request(url, headers, payload, retries=3, delay=5):
+def make_api_request(url, headers, payload, retries=2, delay=3):
     """Make API requests with retry logic."""
     for attempt in range(retries):
         try:
@@ -108,14 +115,13 @@ def make_api_request(url, headers, payload, retries=3, delay=5):
 
 def get_keyword_match(user_input_lower):
     """Find the best matching keyword group using semantic similarity."""
+    init_semantic_model()
     user_embedding = semantic_model.encode(user_input_lower, convert_to_tensor=True)
     best_score = -1
     best_group = None
 
-    for group, keywords in keyword_groups.items():
-        keyword_texts = keywords + [krishna_blessings.get(k, "") for k in keywords if k in krishna_blessings]
-        keyword_embeddings = semantic_model.encode(keyword_texts, convert_to_tensor=True)
-        similarities = util.cos_sim(user_embedding, keyword_embeddings)
+    for group in keyword_embeddings_cache:
+        similarities = util.cos_sim(user_embedding, keyword_embeddings_cache[group])
         max_similarity = similarities.max().item()
         if max_similarity > best_score and max_similarity > 0.6:
             best_score = max_similarity