fix typo

Author: polyzos

website/blog/2026-04-16-fluss-realtime-profiles.md

@@ -35,7 +35,7 @@ the profile is maintained inside Fluss as events arrive, using the Aggregation M
 The stream and the profile are the same system, with the same consistency model and the same recovery guarantees.
 
 This has a meaningful downstream effect: any consumer that can read Fluss gets profiles for free. 
-An OLAP engine like StarRocks or DucksDB can query the profile tables directly, nothing changes about how they work,
+An OLAP engine like StarRocks or DuckDB can query the profile tables directly, nothing changes about how they work,
 they simply no longer have to own the profile state. 
 
 AI applications and feature stores can read the same bitmap-structured groups for real-time feature serving. 
@@ -384,4 +384,4 @@ The three components work together:
 * **Roaring Bitmaps** give compressed, exact-count set representations for entity populations that can reach hundreds of millions of members
 * **The Aggregation Merge Engine** gives those sets a storage-layer home with correct semantics for incremental updates and replay-safe recovery — driven by the Flink connector's UndoRecoveryOperator
 
-The result is a profile infrastructure that answers "what do we know about this entity right now?" with data that is seconds old instead of hours old, applicable wherever events drive decisions, and correct under the failure conditions that production streaming systems encounter every day.
+The result is a profile infrastructure that answers "what do we know about this entity right now?" with data that is seconds old instead of hours old, applicable wherever events drive decisions, and correct under the failure conditions that production streaming systems encounter every day.