There are several examples of work which uses data from so- cial media to detect events which occur in our real, physical world. Our target for event detection is to partition a large geographic region, a whole city in our case, into smaller districts based on geotagged Tweets and to detect smaller local events. We generate a language model for Tweets from each district and measure the KL divergence on incoming Tweets to detect outliers. When these reach a sizable volume or intensity and are consistent, this indicates an event within that district. We used Tweets drawn from Dublin city and we describe experiments on partitioning the city into districts and detecting local events within districts.