diff --git a/server/reflector/hatchet/workflows/topic_chunk_processing.py b/server/reflector/hatchet/workflows/topic_chunk_processing.py
index 1f8e6c47..c199e7a1 100644
--- a/server/reflector/hatchet/workflows/topic_chunk_processing.py
+++ b/server/reflector/hatchet/workflows/topic_chunk_processing.py
@@ -7,12 +7,12 @@ Spawned dynamically by detect_topics via aio_run_many() for parallel processing.
 
 from datetime import timedelta
 
-from hatchet_sdk import Context
+from hatchet_sdk import ConcurrencyExpression, Context
 from hatchet_sdk.rate_limit import RateLimit
 from pydantic import BaseModel
 
 from reflector.hatchet.client import HatchetClientManager
-from reflector.hatchet.constants import LLM_RATE_LIMIT_KEY, TIMEOUT_SHORT
+from reflector.hatchet.constants import LLM_RATE_LIMIT_KEY, TIMEOUT_MEDIUM
 from reflector.hatchet.workflows.models import TopicChunkResult
 from reflector.logger import logger
 from reflector.processors.prompts import TOPIC_PROMPT
@@ -32,12 +32,17 @@ class TopicChunkInput(BaseModel):
 hatchet = HatchetClientManager.get_client()
 
 topic_chunk_workflow = hatchet.workflow(
-    name="TopicChunkProcessing", input_validator=TopicChunkInput
+    name="TopicChunkProcessing",
+    input_validator=TopicChunkInput,
+    concurrency=ConcurrencyExpression(
+        expression="true",  # constant CEL expression = global limit
+        max_runs=20,
+    ),
 )
 
 
 @topic_chunk_workflow.task(
-    execution_timeout=timedelta(seconds=TIMEOUT_SHORT),
+    execution_timeout=timedelta(seconds=TIMEOUT_MEDIUM),
     retries=3,
     rate_limits=[RateLimit(static_key=LLM_RATE_LIMIT_KEY, units=1)],
 )