Astronomer's will soon discover well over 1 million cosmic explosions - known as supernovae - annually. Here, I will describe ongoing work to hunt through this "haystack" of events for the most rare "needles". I will describe probabilistic deep learning methods used to reduce the dimensionality of our data, allowing us to search for out-of-distribution anomalies in real time. Time permitting, I will also discuss recent results utilizing simulation-based-inference for rapid inference of supernova properties.