Update src/core/QuizEngine.py

src/core/QuizEngine.py

@@ -24,27 +24,30 @@ class QuizEngine:
             "question": f"What does **{acronym}** stand for?"
         }
 
-    # --- MODE 2: DOCUMENTS (The Fix) ---
-    def get_document_context(self, username):
+    # --- MODE 2: DOCUMENTS (Updated for Guided Quizzing) ---
+    def get_document_context(self, username, topic_filter=None):
+        """
+        Fetches a context chunk. 
+        If 'topic_filter' is provided, only returns chunks containing that text.
+        """
         user_dir = os.path.join(self.source_dir, username)
         if not os.path.exists(user_dir): return None
 
         files = [f for f in os.listdir(user_dir) if f.lower().endswith(('.txt', '.md'))]
         if not files: return None
 
-        # Try up to 10 times to find a file with valid chunks
-        for attempt in range(10):
+        # Shuffle files to ensure randomness even when filtering
+        random.shuffle(files)
+
+        # Try up to 20 times (increased from 10 to handle filtering)
+        for attempt in range(20):
             selected_file = random.choice(files)
             try:
                 file_path = os.path.join(user_dir, selected_file)
                 with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
                     text = f.read()
 
-                # --- NEW CHUNKING STRATEGY: SLIDING WINDOW ---
-                # Instead of splitting by paragraphs (which kills lists),
-                # we grab chunks of roughly 1000-1500 characters.
-                # This ensures we get the Header + The Bullets together.
-                
+                # --- SLIDING WINDOW STRATEGY ---
                 step_size = 1000
                 window_size = 1500
                 candidates = []
@@ -53,18 +56,19 @@ class QuizEngine:
                 if len(text) < window_size:
                     candidates.append(text)
                 else:
-                    # Slide a window across the text
                     for i in range(0, len(text) - window_size, step_size):
                         chunk = text[i : i + window_size]
                         
-                        # BASIC SANITY CHECKS (Very Relaxed)
-                        # 1. Must have content
+                        # 1. Sanity Check
                         if len(chunk.strip()) < 100: continue
-                        
-                        # 2. Must not be pure administrative junk
-                        # We let the LLM decide mostly, but we filter obvious "Blank Pages"
                         if "intentionally left blank" in chunk.lower(): continue
                         
+                        # 2. TOPIC FILTER (The New Logic)
+                        if topic_filter:
+                            # Case-insensitive search
+                            if topic_filter.lower() not in chunk.lower():
+                                continue
+
                         candidates.append(chunk)
 
                 if not candidates: continue
@@ -85,24 +89,19 @@ class QuizEngine:
         return None
 
     def construct_question_generation_prompt(self, context_text):
-            """
-            REVISED PROMPT: 'Action-Based' Generation.
-            Forces the model to generate a question for ANY valid information, 
-            lowering the bar for what constitutes 'examinable'.
-            """
-            return (
-                f"Act as a Navy Board Examiner.\n"
-                f"Here is a raw text excerpt from Navy documentation:\n"
-                f"'''{context_text}'''\n\n"
-                f"TASK: Generate a single question based on this text.\n\n"
-                f"RULES:\n"
-                f"1. If the text contains ANY facts (dates, definitions, locations, process steps, responsibilities), you MUST generate a question.\n"
-                f"2. Do not judge the 'quality' of the fact. If the text says 'The NSWC is in Maryland', ask 'Where is the NSWC located?'.\n"
-                f"3. Only output 'UNABLE' if the text is completely meaningless (e.g., a string of page numbers, a blank page, or garbled characters).\n\n"
-                f"OUTPUT FORMAT:\n"
-                f"- If valid: Just the question text.\n"
-                f"- If invalid: The word 'UNABLE'."
-            )
+        return (
+            f"Act as a Navy Board Examiner.\n"
+            f"Here is a raw text excerpt from Navy documentation:\n"
+            f"'''{context_text}'''\n\n"
+            f"TASK: Generate a single question based on this text.\n\n"
+            f"RULES:\n"
+            f"1. If the text contains ANY facts (dates, definitions, locations, process steps, responsibilities), you MUST generate a question.\n"
+            f"2. Do not judge the 'quality' of the fact. If the text says 'The NSWC is in Maryland', ask 'Where is the NSWC located?'.\n"
+            f"3. Only output 'UNABLE' if the text is completely meaningless (e.g., a string of page numbers, a blank page, or garbled characters).\n\n"
+            f"OUTPUT FORMAT:\n"
+            f"- If valid: Just the question text.\n"
+            f"- If invalid: The word 'UNABLE'."
+        )
 
     def construct_grading_prompt(self, question, answer, context_text):
         return (