Building a machine translation using Moses is quite an involved process that involves tokenization, language model training, data cleaning, training, tuning, testing, evaluation, etc. We have developed a program called experiment.perl that allows you to specify a configuration file and the program figures out what to run (and in which order), schedules jobs on clusters or forks them out, etc. The goal of this project would be to extend this program to incorporate more open source tools, and overcome some of its limitations.