Write up your GSOC idea for Mahout. Make sure it is well referenced and well written. Do not take on the world in your proposal. In my experience (I've done it 2 years now) most students have time to implement 1 algorithm + documentation + examples + unit tests. That is it. I pretty much reject any proposal that says differently, as it simply implies you don't get what is involved in the task. There are lots of opportunities out there for adding capabilities to Mahout.
See the Mahout mail archives in past years for tips: http://www.lucidimagination.com/search/?q=gsoc+Mahout