I would not focus his time on black box HFT, as that really involves computer infrastructure more than quant work. I would ask him to develop a futures trading model based on yield curve spreads. This is quantitative, interesting, and has lots of potential if he is really a genius.
1. Assign him a "project" to illustrate a curve fit: a. using TA of his choice (macd crossover or something basic) create a directional trading program and optimize over: i. one month ii. six months iii. one year iv. five years v. 10 years of data b. Each trial will include an out of sample (blind) performance report at least 20% as long as the training window c. Conclude the assignment with a graph of OOS performance vs. in sample returns for the various time frames 2. Repeat the assignment using another TA, fundamental, spread, correlated, or other strategy. 3. Present the most robust (least degradation OOS as % of in sample) and the highest performing (max return as % of max DD) 4. Conduct sensitivity study on these two finalists by tweaking the parameterz and noting the performance change ...... that should keep him going for a week or two.