How should a firm allocate its limited interviewing resources to select theoptimal cohort of new employees from a large set of job applicants? How shouldthat firm allocate cheap but noisy resume screenings and expensive but in-depthin-person interviews? We view this problem through the lens of combinatorialpure exploration (CPE) in the multi-armed bandit setting, where a centrallearning agent performs costly exploration of a set of arms before selecting afinal subset with some combinatorial structure. We generalize a recent CPEalgorithm to the setting where arm pulls can have different costs and returndifferent levels of information. We then prove theoretical upper bounds for ageneral class of arm-pulling strategies in this new setting. We apply ourgeneral algorithm to a real-world problem with combinatorial structure:incorporating diversity into university admissions. We take real data fromadmissions at one of the largest US-based computer science graduate programsand show that a simulation of our algorithm produces a cohort with hiringoverall utility while spending comparable budget to the current admissionsprocess at that university.