Refined Logic

Added chat history as a component in consideration when assigning the mode to respond with.

app.py

@@ -114,8 +114,17 @@ Be concise and direct with an overall friendly and engaging tone. Use minimal fo
 ])
 
 # --- Core Logic Functions ---
-def detect_subject(message):
-    """Detects the subject of the user's message based on keywords."""
+def detect_subject(message, history=None):
+    """
+    Detects the subject of the user's message based on keywords and conversation context.
+    
+    Args:
+        message (str): Current user message
+        history (list): Chat history with role/content structure
+    
+    Returns:
+        tuple: (template, mode_string)
+    """
     message_lower = message.lower()
     
     math_keywords = ['math', 'mathematics', 'solve', 'calculate', 'equation', 'formula', 'algebra', 'geometry', 'calculus', 'derivative', 'integral', 'theorem', 'proof', 'trigonometry', 'statistics', 'probability', 'arithmetic', 'fraction', 'decimal', 'percentage', 'graph', 'function', 'polynomial', 'logarithm', 'exponential', 'matrix', 'vector', 'limit', 'differential', 'optimization', 'summation']
@@ -124,28 +133,84 @@ def detect_subject(message):
     
     study_keywords = ['study', 'studying', 'memorize', 'memory', 'exam', 'test', 'testing', 'quiz', 'quizzing', 'review', 'reviewing', 'learn', 'learning', 'remember', 'recall', 'focus', 'concentration', 'motivation', 'notes', 'note-taking', 'flashcard', 'flashcards', 'comprehension', 'understanding', 'retention', 'practice', 'drill', 'preparation', 'revision', 'cramming']
     
-    if any(keyword in message_lower for keyword in math_keywords):
+    # Score the current message
+    current_scores = {
+        'math': sum(1 for keyword in math_keywords if keyword in message_lower),
+        'research': sum(1 for keyword in research_keywords if keyword in message_lower),
+        'study': sum(1 for keyword in study_keywords if keyword in message_lower)
+    }
+    
+    # Analyze recent conversation history for context
+    context_scores = {'math': 0, 'research': 0, 'study': 0}
+    last_mode = None
+    
+    if history:
+        # Look at last 3-5 exchanges for context
+        recent_history = history[-6:]  # Last 3 user-bot exchanges
+        
+        for i, exchange in enumerate(recent_history):
+            if exchange.get("role") == "user":
+                content_lower = exchange.get("content", "").lower()
+                
+                # Weight recent messages more heavily
+                weight = 0.5 - (i * 0.1)  # More recent = higher weight
+                if weight < 0.1:
+                    weight = 0.1
+                
+                context_scores['math'] += weight * sum(1 for keyword in math_keywords if keyword in content_lower)
+                context_scores['research'] += weight * sum(1 for keyword in research_keywords if keyword in content_lower)
+                context_scores['study'] += weight * sum(1 for keyword in study_keywords if keyword in content_lower)
+            
+            # Check if bot responses indicate a specific mode
+            elif exchange.get("role") == "assistant":
+                content = exchange.get("content", "")
+                if "*Math Mode*" in content:
+                    last_mode = 'math'
+                elif "*Research Mode*" in content:
+                    last_mode = 'research'
+                elif "*Study Mode*" in content:
+                    last_mode = 'study'
+    
+    # Combine current message scores with context
+    final_scores = {
+        'math': current_scores['math'] * 2.0 + context_scores['math'],  # Current message weighted more
+        'research': current_scores['research'] * 2.0 + context_scores['research'],
+        'study': current_scores['study'] * 2.0 + context_scores['study']
+    }
+    
+    # Add bonus for continuing in the same mode (conversation persistence)
+    if last_mode and last_mode in final_scores:
+        final_scores[last_mode] += 0.5
+    
+    # Handle ambiguous/continuation messages
+    ambiguous_phrases = ['next', 'continue', 'more', 'explain', 'help', 'what about', 'can you', 'how do', 'show me']
+    is_ambiguous = any(phrase in message_lower for phrase in ambiguous_phrases) and len(message.split()) < 6
+    
+    if is_ambiguous and last_mode:
+        # For short, ambiguous messages, stick with the last mode
+        if last_mode == 'math':
+            return math_template, "Math Mode"
+        elif last_mode == 'research':
+            return research_template, "Research Mode"
+        elif last_mode == 'study':
+            return study_template, "Study Mode"
+    
+    # Determine the best mode based on scores
+    max_score = max(final_scores.values())
+    
+    if max_score == 0:
+        return general_template, "General Mode"
+    
+    # Return the mode with the highest score
+    if final_scores['math'] == max_score:
         return math_template, "Math Mode"
-    elif any(keyword in message_lower for keyword in research_keywords):
+    elif final_scores['research'] == max_score:
         return research_template, "Research Mode"
-    elif any(keyword in message_lower for keyword in study_keywords):
+    elif final_scores['study'] == max_score:
         return study_template, "Study Mode"
     else:
         return general_template, "General Mode"
 
-def smart_truncate(text, max_length=3000):
-    """Truncates text intelligently to the last full sentence or word."""
-    if len(text) <= max_length:
-        return text
-    
-    # Try to split by sentence
-    sentences = re.split(r'(?<=[.!?])\s+', text[:max_length])
-    if len(sentences) > 1:
-        return ' '.join(sentences[:-1]) + "... [Response truncated - ask for continuation]"
-    # Otherwise, split by word
-    else:
-        words = text[:max_length].split()
-        return ' '.join(words[:-1]) + "... [Response truncated - ask for continuation]"
 
 def respond_with_enhanced_streaming(message, history):
     """Streams the bot's response, detecting the subject and handling errors."""
@@ -158,7 +223,8 @@ def respond_with_enhanced_streaming(message, history):
     mode = ""
     
     try:
-        template, mode = detect_subject(message)
+        # UPDATED: Pass history to detect_subject
+        template, mode = detect_subject(message, history)
         
         # Build conversation history with proper LangChain message objects
         messages = []