Storm Trident group by selected fields before windowing

Question

I have a simple case study:

I want to use Trident Storm to calculate the average temperature from multiple sensors. Kafka is used as the source and sink data. The data from one sensor looks like this:

{
  "uuid": "e5778f99-d300-42cb-9992-cf453aa2d25f",
  "sensor_uuid": "e5778f99-d400-42cb-9992-cf453272d25f",
  "temp": "24.8"
}

Is there a way in Trident Storm to group events (by the field named sensor_uuid), then apply windowing, which allows calculation?

My current code:

this.topologyBuilder
   .newStream("kafkaSpout", createKafkaTridentSpoutOpaque())
   .tumblingWindow(12, new InMemoryWindowsStoreFactory(), new Fields("value"), new AggEvents(), new Fields("message"))
   .partitionPersist(createTridentKafkaStateFactory(), new Fields("message"), new TridentKafkaStateUpdater<String, String>(), new Fields());

Unfortunately, I can't use groupBy() before tumblingWindow() because it returns GroupedStream, which doesn't support windowing.

How can I achieve my goal?