AESTRA (Auto-Encoding STellar Radial-velocity and Activity) uses deep learning for precise radial velocity measurements in the presence of stellar activity noise. The architecture combines a convolutional radial-velocity estimator and a spectrum auto-encoder. The input consists of a collection of hundreds or more of spectra of a single star, which span a variety of activity states and orbital motion phases of any potential planets. AESTRA does not require any prior knowledge about the star.