Understanding the properties of our dark matter halo is relevant to both astrophysics as it informs the formation history of our galaxy, and particle physics in that it impacts the interpretation of dark matter experiments. This talk reviews the spherical cow assumptions that underly the model for the halo that is typically assumed, and then questions those assumptions, making clear that a data driven approach is warranted. Recent work that has shown that low metallicity stars can act as tracers for dark matter, which has allowed data from the Gaia satellite to be interpreted as a measurement of the halo. However, the number of stars in the Gaia dataset with metallicity information (~200,000) is a small fraction of the order 1 billion measured stars. With the aid of modern machine learning technology, we seek to find if the tracer stars can be identified only by the kinematic information measure by Gaia. Our positive initial results indicate that it could be possible to “learn the dark matter halo” with much finer resolution than is currently possible.