This paper presents a smart driving direction system that leverages taxi drivers' intelligence and experience. GPS-equipped taxis act as mobile sensors to model the dynamic traffic patterns of a city. A time-dependent landmark graph models traffic patterns and experienced drivers' route choices to provide users with the practically fastest route to a destination at a given departure time. A clustering approach estimates travel times between landmarks in different time slots. A two-stage routing algorithm then computes a customized, practically fastest route for the user based on this graph and real-world trajectory data from over 33,000 taxis over three months. Evaluation found that 60-70% of routes suggested were faster than alternatives, with 20% sharing results, and on average routes