This thesis describes a search for long-lived heavy neutral leptons using a dataset of 137/fb collected during the 2016-2018 proton-proton runs with the CMS detector. The search uses a final state containing two leptons and at least one hadronic jet. This is the first analysis at the Large Hadron Collider which considers universal mixing between the Standard Model and heavy neutral lepton species. The search makes heavy use of a deep neural network-based displaced jet tagging algorithm, originally developed to target heavy long-lived gluino decays. The tagger was trained on both simulation and proton-proton collision data using the domain adaptation technique, which significantly improved the modelling of its output in simulation. The tagger has excellent performance for a range of long-lived particle lifetimes and generalises well to various flavours of displaced jets. In this analysis, the backgrounds are estimated in an entirely data-driven manner. No evidence for heavy neutral leptons is observed, and upper limits are set for a wide range of heavy neutral lepton mass, lifetime, and mixing scenarios. This is the most sensitive search for heavy neutral leptons in the 1–12 GeV mass range to date.