The concept behind the site is that while you're travelling, the people who might have the most relevant information for you aren't necessarily the people you follow, but rather the local people on Twitter who already know the area. Twaller's goal is to connect you with those people and the tweets they're sending that are relevant to you in your current location.
The site uses a keyword-based search engine and natural language processing to identify tweets related to at present 5 major buckets in major metro areas: weather, places to eat, things to see, entertainment, and shopping. When you visit the site you can look at all recent travel-related tweets or browse tweets by the city you're in.
Once you browse into a city, tweets are organized by the 5 topics mentioned above. For the traveller, it gives an interesting zeitgeist of tweets that might matter to you from sources you're not necessarily following, but have high relevancy due to location and context.
If you sign in to the site with your Twitter credentials using OAuth, you can post your own tweets and categorize them in any of the 5 travel-related categories. You can also retweet any of the tweets you see on the site if you find them useful or interesting to other travelers or followers.
Twaller is a great first step towards building a more context-aware filtering system on top of Twitter, in this case specifically for travellers, bi-coasters and other multi-location individuals. Amitabha Banerjee, co-founder of Twaller, admitted as much in his presentation that once Twitter's location functionality goes live any day now, the service will be able to skyrocket in its utility -- solving current problems such as ambiguity in city and restaurant or location names, for one thing.