This is a demo of Infer.NET which is a .NET library for inference in graphical models. Unlike black-box machine learning packages, Infer.NET allows you to piece together complex probabilistic models via a straightforward programming interface. Compared to existing inference tools like BUGS and VIBES, Infer.NET is unique in that it can perform either Expectation Propagation or Variational Message Passing on models with discrete and continuous variables. It supports arbitrary mixture models via factor graphs with gates. It is also uniquely structured as a compiler, taking a .NET program with random variables as input and producing a specially-optimized inference program on output.