For the past few months I've been teaching myself Longhorn programming via the PDC bits by building a better Spam filter. Some of the best filters will typically use an approach based on Bayesian probability modeling. For example, in this article Paul Graham discusses an improved algorithm for filtering spam
Although I am currently a software engineer (and manager) by trade, my background is in Mathematics and I'm not entirely satisfied with the current crop of spam filters. I realized that mathematical analysis alone isn't enough. There has to be an "intelligence" factor.
The app I'm working on takes the Bayesian approach, but integrates that with intelligent filtering by taking advantage of the cheap cost of outsourcing and the proliferation of peer-to-peer software (made easier by Indigo I might add). You can see my new spam filter here.