This work presents a digital twin framework designed to support radiofrequency (RF) ablation procedures by integrating synthetic electrophysiological simulations and machine learning (ML)-based electrogram (EGM) classification. The system aims to aid clinicians in identifying arrhythmogenic substrates with in-silico modeling. The digital twin is implemented using cardiac data extracted from Late Gadolinium Enhancement Magnetic Resonance Imaging (LGE-MRI) using the openCARP simulation platform that models atrial tissue activity. EGMs are synthetically generated by navigating a virtual catheter across atrial tissue. A supervised ML classifier is trained on these synthetic signals to classify ablation sites based on the EGM signal. Ablation is emulated by modifying tissue conductivity and anisotropy at selected points, governed by RF parameters such as duration and power. The system dynamically updates its electrophysiological state to reflect propagation changes due to lesion formation. The developed system aims to enable effective monitoring of the ablation procedure through a classifier and ablation quality index (AQI), providing visual results that can assist clinicians in decision-making before, during, or after the ablation procedure. This digital twin aims at offering a versatile and realistic platform for studying ablation strategies and training ML models on synthetic EGMs. It provides a non-invasive, controllable, and reproducible environment to test hypotheses in cardiac ablation, with potential for real-time clinical decision support.