diff --git a/generator/src/llm_utils.py b/generator/src/llm_utils.py
index 7ed1d29..7e54c9e 100644
--- a/generator/src/llm_utils.py
+++ b/generator/src/llm_utils.py
@@ -1,14 +1,19 @@
 import requests
 import logging
+import asyncio
+from functools import partial
 from generator.src.config import OLLAMA_MODEL
 
 logger = logging.getLogger(__name__)
 
+# 🔥 Limit concurrency
+semaphore = asyncio.Semaphore(5)
+
+
 def generate_text(prompt, model=None, url_override=None):
     """Generate text from prompt using local Ollama."""
     target_model = model or OLLAMA_MODEL
-    
-    # Use centralized config unless overridden
+
     if url_override:
         url = url_override
     else:
@@ -24,27 +29,28 @@ def generate_text(prompt, model=None, url_override=None):
             "num_ctx": 8192
         }
     }
-    
-    logger.info(f"Sending request to Ollama (model: {target_model}, prompt length: {len(prompt)} chars)")
-    
+
     try:
         resp = requests.post(url, json=payload, timeout=600)
         resp.raise_for_status()
-    except Exception as e:
-        logger.error(f"Error calling Ollama at {url}: {e}", exc_info=True)
-        return ""
-
-    try:
         data = resp.json()
-        response_text = data.get("response", "")
-        
-        logger.info(f"Received response from Ollama (length: {len(response_text)} chars)")
-        logger.info(f"Response done: {data.get('done', False)}")
-        
-        if not response_text:
-            logger.warning(f"Empty response from Ollama. Full data: {data}")
-        
-        return response_text
+        return data.get("response", "")
     except Exception as e:
-        logger.error(f"Error parsing Ollama response: {e}", exc_info=True)
+        logger.error(f"LLM error: {e}", exc_info=True)
         return ""
+
+
+async def generate_text_async(prompt, model=None, url_override=None):
+    loop = asyncio.get_event_loop()
+    func = partial(generate_text, prompt, model, url_override)
+    return await loop.run_in_executor(None, func)
+
+
+async def controlled_generate_text(prompt):
+    async with semaphore:
+        return await generate_text_async(prompt)
+
+
+async def process_batch(prompts):
+    tasks = [controlled_generate_text(p) for p in prompts]
+    return await asyncio.gather(*tasks, return_exceptions=True)