Following the discussion in #324 , here a project proposal for GPflowOpt. While I was listing the responses to the questions raised in order to describe the project, it got a bit long so I made a document (attached) describing the goals, and comparison to existing alternatives.
gpflowopt.pdf.zip
Internally, we had some code available for BO with GPflow, I took parts of it and put up a small demo with a notebook containing a constrained EI example (https://github.com/javdrher/optdemo). We would start based on this, test it fully and initiate proper documentation, rather than the brief comments now. Only then expand further to increase the number of supported features such as more complex domains (and optimizers), better designs, more acquisition functions etc.

Of course, we are available to discuss this on Github or over Skype. Additionally I can head over to the UK for a face-to-face discussion, should that be helpful.