Gave Perplexity Max a single task: predict CFPB enforcement actions from public complaint data. Told it nothing about BoTorch — it found the repo, chose MixedSingleTaskGP, ran 48 BO vs 48 random evals. BO mean F1=0.725 vs 0.389. Total cost: $200. The model is now scoring live companies. Happy to answer questions about the methodology or the BO search space design.