In recent years, survey facilities have started to provide remarkable new opportunities for transient astronomy. This is made possible by combining unprecedented fields of view with high sensitivity in uninvestigated wavelength regimes. The result of this combination is an extremely high volume of data to be analysed. In this work, we have opened up the way to real-time automated analysis of the image data-stream. We present a fully automated GPU-based machine-learning backed pipeline for analysis of radio images at multiple frequencies. It consists of four consecutive steps: quality control, source detection, association, flux measurement and physical parameter inference. At the end of the pipeline, an alert of a significant detection can be sent out and data will be saved for further investigation.In our current implementation, the entire pipeline analyses images from AARTFAAC (Amsterdam Astron Radio Transients Facility And Analysis Centre) with sizes upwards of 16 × 1024 × 1024 pixels. AARTFAAC is a transient detector based on the Low-Frequency Array, an interferometer based in the Netherlands with stations across Europe. First results show that dispersed signals were found on which follow-up analysis can be performed. The identification and response to transients is the key science goal for AARTFAAC, and the current work brings us one step closer to meeting it.