Vpon is a leading mobile advertising company operating in the APAC region, currently we process around 40 million Ad requests daily. In order to deliver better suited Ads to our audience, we have developed an "item-based" recommendation system to help decide who might be interested in a certain campaign. In order to process this huge amount of data in a daily basis, we have based this system on Hadoop and picked Python for fast development. In this talk, we are going to introduce how we implemented this system using Python with Hadoop Streaming. We will start by explaining the background of this work, then go into the data we use to feed into this system, the actual implementation and how it interacts with the our Ad delivery system.