Lukas Steinwender
Lukas Steinwender is a PhD Candidate at Swinburne University of Technology Australia. He initially completed a Masters in Physics at the University of Graz, focusing on astrophysics of RR Lyrae stars. Along the way he developed a strong passion for computer-science which led to a second degree completed at Graz University of Technology. Combining both of these passions in a PhD, he is exploring modern machine learning methods and developing visualization tools. His research aims to apply those methods to better understand the universe using some of the largest datasets in optical astronomy, including the Vera C. Rubin Observatory Legacy Survey of Space and Time.
Session
The now operational Rubin LSST monitors the southern night sky and is forecast to send 10 million alerts (objects with significant changes in brightness) per night to brokers such as Fink. Fink is tasked with processing and filtering the alert-stream to manageable amounts for the individual researcher. During the first period of public Rubin alerts (February 2026), over 800 thousand alerts were streamed through the pipeline. Discovering anomalies and assessing image quality at scale in the real-time data streams of the Fink ecosystem is challenging.
ThumP! (Thumbnails? Prescreen!) was planned, implemented, and deployed successfully to visualize the first public alerts streaming through Fink providing confidence in the nominal operation and performance of the system. In the process ThumP! aided the discovery of 50 new supernova candidates and helped evaluate stream filters for good detections. This was achieved through ThumP!’s fast, dynamic visualization of Rubin thumbnails allowing the simultaneous (multi-user), parallel inspection and comparison of a large number (30 per minute) of candidate images. Individual visualizations can be adjusted interactively for easier comparison, annotated for quality control, and exported for further inspection or triggering follow-up with other telescopes. During the first public alerts stream ThumP! was deployed on a large-format tiled display wall, allowing the local Fink team to inspect 12 000 alerts in a couple of hours.
ThumP! is open source and publicly accessible on the web.