In this paper we study the problem of guessing what search query the user intends to type into a search engine based on the first few characters of the query, also known as prefix based query auto-completion. We train and evaluate two personalized auto-completion models on search logs from an online news portal. The personalization comes from using demographic and location information specific to the user. Our experiments show that we can guess the query the user intended to type and rank it among the top three suggestions over 75 % of the time. Moreover, the methods described can decrease the number of keystrokes by about 40%, thus saving the user a lot of typing.