In the infinite data and compute limit, machine learning (ML) methods can be optimal, however this idealistic situation is not often realized in practice. On the other hand, principled data-analysis methods are robust, but they make simplistic assumptions (e.g., the noise is roughly Gaussian). I will present how ML algorithms can enhance matched-filtering gravitational wave pipelines by generating optimally-compressed template banks and mitigating non-Gaussian noise. Incorporating these advancements in the IAS gravitational-wave search pipeline, I will present new detections of black holes in the astrophysically significant pair-instability mass gap and intermediate-mass black hole (IMBH) ranges. Towards the end, I will change gears and explore applications of interpretable AI tools in compact object formation channels, galaxy formation, and cosmology.