Update app.py

app.py

@@ -32,20 +32,19 @@ def socratic_streaming_chat(user_query):
         "-p", formatted_chat_prompt,
         "-n", "120",
         "-temp", "0.4",
-        "-t", "2"  # Optimized for Hugging Face free-tier dual-core CPUs
+        "-t", "2"
     ]
     
-    # Spawn the process with stdout piping enabled for live stream reading
     process = subprocess.Popen(
         cmd, 
         stdout=subprocess.PIPE, 
-        stderr=subprocess.DEVNULL,  
+        stderr=subprocess.DEVNULL,  # Keeps system logs hidden
         text=True, 
         bufsize=1
     )
     
     accumulator = ""
-    # Lookahead buffer size in characters (~45 chars safely covers variance of stop loops)
+    prompt_cleared = False
     LOOKAHEAD_SIZE = 45 
     
     stop_markers = [
@@ -55,19 +54,22 @@ def socratic_streaming_chat(user_query):
         "Assistant:"
     ]
     
-    # Read the terminal execution stream live character-by-character
     while True:
         char = process.stdout.read(1)
         if not char:
-            break  # End of stream reached
+            break 
             
         accumulator += char
         
-        # Discard the echoed prompt wrapper if it slips into the stdout read window
-        if accumulator.startswith(formatted_chat_prompt):
-            accumulator = accumulator.replace(formatted_chat_prompt, "").strip()
+        # --- THE FIX: SWALLOW THE ECHOED PROMPT ---
+        if not prompt_cleared:
+            if "Assistant:" in accumulator:
+                prompt_cleared = True
+                # Delete the prompt and keep only what comes after "Assistant:"
+                accumulator = accumulator.split("Assistant:")[-1].lstrip()
+            continue  # Do not yield text until the prompt is fully swallowed
             
-        # Scan the live accumulator for structural collapse boundaries
+        # Scan for structural collapse boundaries
         stop_triggered = False
         for marker in stop_markers:
             if marker in accumulator:
@@ -76,15 +78,14 @@ def socratic_streaming_chat(user_query):
                 break
                 
         if stop_triggered:
-            process.terminate()  # Hard-kill the engine to stop burning CPU cycles
+            process.terminate() 
             break
             
-        # UX Lookahead Protection: Only stream text that sits safely outside the trailing boundary window
+        # Stream text safely outside the trailing boundary window
         if len(accumulator) > LOOKAHEAD_SIZE:
             safe_to_display = accumulator[:len(accumulator) - LOOKAHEAD_SIZE]
             yield safe_to_display.strip()
             
-    # Yield the completely un-buffered, finalized clean string 
     yield accumulator.strip()
 
 # ==============================================================================