Online Supernova pointing with Trigger Primitives in DUNE

We present a fast, DAQ-level pipeline to infer the direction of a core-collapse supernova (SN) using only Trigger Primitives (TPs) in the Deep Underground Neutrino Experiment (DUNE). The approach clusters TP hits, identifies electron tracks, tags interaction channels, and regresses electron directions with convolutional neural networks. A final sky direction is obtained by sampling a likelihood over track directions of Elastic Scattering interactions. This quasi-real-time method is designed to deliver a pointing estimate to SNEWS within minutes after a burst while raw waveforms are still being transferred. We report the performances of the individual steps of the analysis and preliminary pointing performances, which be reassessed as simulation and reconstruction developments continue.