This work presents a pipeline for generating realistic synthetic electrograms (EGMs) simulating sustained atrial fibrillation (AF) using openCARP. Patient-specific cardiac meshes integrating fibrosis and fiber orientation, extracted from Late Gadolinium Enhancement Magnetic Resonance Imaging (LGE-MRI), enable high-fidelity datasets for downstream tasks, such as diagnostic and machine learning applications, complementing real-world clinical data. A protocol is built to guarantee simulation stability using different simulation resolutions, using the Courtemanche cell model with ionic modifications to replicate the electrophysiological behavior of AF. Sustained AF is induced through multiple Archimedean spiral wavefronts, generating persistent reentrant activity. A virtual catheter, randomly positioned on the atrial surface, collects synthetic bipolar EGMs. The resulting signals exhibit morphological features consistent with clinical AF. In a blinded evaluation with five expert electrophysiologists, two experts could not reliably distinguish real from synthetic EGMs, while three reported detecting patterns suggestive of synthetic origin. These results indicate that, although morphology aligns with clinical data, there are patterns in the synthetic data that warrant further refinement to enhance realism. This framework provides a reproducible methodology that addresses data scarcity and supports the development of personalized AF treatment and EGM analysis tools.