The goal of the search for extraterrestrial intelligence (SETI) is to quantify the prevalence of technological life beyond Earth via their ‘technosignatures’. One theorized technosignature is narrowband Doppler drifting radio signals. The principal challenge in conducting SETI in the radio domain is developing a generalized technique to reject human radiofrequency interference. Here we present a comprehensive deep-learning-based technosignature search on 820 stellar targets from the Hipparcos catalogue, totalling over 480 h of on-sky data taken with the Robert C. Byrd Green Bank Telescope as part of the Breakthrough Listen initiative. We implement a novel β-convolutional variational autoencoder to identify technosignature candidates in a semi-unsupervised manner while keeping the false-positive rate manageably low, reducing the number of candidate signals by approximately two orders of magnitude compared with previous analyses on the same dataset. Our work also returned eight promising extraterrestrial intelligence signals of interest not previously identified. Re-observations on these targets have so far not resulted in re-detections of signals with similar morphology. This machine-learning approach presents itself as a leading solution in accelerating SETI and other transient research into the age of data-driven astronomy.