Background Atrial fibrillation (AF) affects 38 million individuals worldwide and is responsible for $26 billion in healthcare costs in the U.S. alone, with 75% of costs attributed to hospitalizations. Early rhythm control improves cardiovascular outcomes in AF and is a cornerstone of treatment. Class III antiarrhythmic drugs (AADs) (e.g., dofetilide, sotalol) typically require inpatient initiation due to the risk of QT prolongation and torsades de pointes (TdP). This limits access and increases healthcare costs, with a single 3-day hospitalization for AAD initiation costing approximately $13,400. This burden disproportionately affects underserved populations, including minorities and rural populations, who may have limited access to appropriate healthcare facilities. We hypothesized that a novel machine learning (ML) algorithm paired with the SafeBeat Rx technology could enable safe outpatient AAD initiation, potentially improving access to care and reducing costs. Methods We developed CNN-BiLSTN based ML-model to automate ECG analysis and QTc measurement to drive AAD dosing recommendations. The model was trained on approximately 9M hand-annotated ECG heartbeats, utilizing data augmentation (e.g. noise) and k-fold cross validation (k=5). We then conducted a prospective study of patients admitted for in-patient dofetilide or sotalol initiation. Parallel to standard-of-care drug dosing, based on physician measured QTc on 12-lead ECG at baseline and 2 hours after each dose, patients were instructed to simultaneously record a mECG device (KardiaMobile 6L; AliveCor) which was transmitted to the software application. Software performance was evaluated by comparing software-annotated, physician adjudicated mECG QTc to blinded physician-measured 12-lead ECG QTc. Results A total of 50 patients (mean age 67 + 11 years, 64% male, 64% caucasian, 22% Asian/Pacific Islander, 8% Hispanic, mean baseline creatinine clearance 85.5 + 33.0mL/min, mean baseline QTc 417.7 + 30.1ms) underwent AAD initiation (44 dofetilide, 6 sotalol). The ML-based QTc measurement algorithm was used to analyze 314 mECGs that were used to guide 264 AAD doses. The ML- and physician-prescribed AAD dosing were congruent or more conservative in 245/264 (92.8%). Conclusion Overall, the software-recommended AAD drug dosing utilizing the ML QTc measurement was in high agreement with cardiologist-determined AAD drug dosing. These data demonstrate this software platform's potential to match clinician decisions for AAD dosing, offering the ability to support streamlined AAD titration regardless of patient setting. This approach shows promise for enabling safe outpatient initiation of AADs, potentially improving access to antiarrhythmic treatment and reducing healthcare costs.