We use an automated processing procedure to derive a new catalog of earthquake origins by analyzing approximately 5 years of data from 2014 through early 2019 in the Timor-Leste and Eastern Indonesia region, where complex transition from oceanic lithosphere subduction to arc-continental collision is actively occurring. The data are from a temporary deployment of 30 broadband seismometers across this region (Miller et al., 2016) and three permanent stations (GE network GEOFON Data Centre, 1993). Deep-neural-network-based phase picker, EQTransformer (Mousavi et al., 2020) and a sequential earthquake association and location workflow, including REAL (Zhang et al., 2019), PyKonal (White et al., 2020) and HypoDD (Waldhauser and Ellsworth, 2000) were used in the analyses. We detected and relocated about 20,000 events, ~6 times more compared to the International Seismological Centre (ISC) reported events during this time. The obtained catalogue shows spatiotemporal seismicity patterns across the region with ML ranging from 0.5 to 7, that extend from the surface down to 640 km depth. Our study demonstrates that it is possible to characterize earthquake sequences from raw seismic data using a welltrained machinelearning picker for a complex convergent plate setting. The results of this study provide the most complete catalogue available for the region for the duration of the temporary deployment, which includes a complex pattern of crustal events across the collision zone and into the backarc, as well as abundant deep slab seismicity.