ATK-SE has parameters for the III-V materials, but one of the major difficulties is that often one is interested not in simple AlAs or GaAs, but Al(0.3)Ga(0.7)As. In this case, the brute-force approach requires you to have 10 As atoms and 10 atoms where you randomly set 3 to Al and 7 to Ga. But now you have to do that, more or less, for all original atoms, so the system grows by 10x in atom count. Plus you have to try many different random combinations, and don't get started thinking about Al(0.31)Ga(0.29)As!
So it's a bit of a challenge to study "usual" III-V materials and devices from an atomistic perspective. Of course, as long as you stick to the binary materials you are fine, and one can use this perhaps as a tool to derive parameters for other models, like k.p.
One approach which I have been looking into a bit is called SQS. This may be doable on the Python level, but we don't have anything ready at the moment. Also the virtual crystal method could in future become part of ATK, but it's not implemented yet.
Thus, ultimately the answer depends a lot on what your colleague is looking to do, precisely.