Introduction: The clinical management of smoldering multiple myeloma (SMM), a precursor condition to multiple myeloma (MM), hinges on accurately predicting its highly variable progression to active MM. While static models provide a baseline risk score, they fail to incorporate the dynamic nature of a patient’s disease trajectory. The primary mathematical challenge lies in effectively modeling irregularly sampled, variable-length clinical time-series data within a survival analysis framework. Traditional survival models, such as the Cox Proportional Hazards model, are not designed to handle sequential inputs, while standard deep learning models do not naturally account for right-censored survival data. Methods: To address this, we developed a novel deep survival learning system, CALM: Cancer AI Longitudinal Modeling. We retrospectively analyzed clinical data from 438 patients at MSKCC diagnosed with SMM between 2002 and 2019, with a median follow-up of 4.24 years and a median of 11 laboratory assessments per patient (median interval between measurements: 92 days). At the last follow-up, 185 patients (42.2%) progressed to MM, while 253 (57.8%) were censored. CALM utilizes time-series measurements of key biomarkers, including serum M-protein and free light chains, along with 35 additional laboratory values. Mechanistically, CALM leverages a padded Long Short-Term Memory (LSTM) network with an attention mechanism using a Cox Proportional Hazards loss function designed to recognize patterns and trends across sequential data. The LSTM processes each patient’s longitudinal data as a sequence, learning to identify temporal changes that may signal impending progression. To translate the model’s output into a clinical tool, CALM generates personalized dashboards with risk scores predicted at 3-month intervals, using only the data available up to each time point. Results: Risk scores computed achieved a mean concordance index of 0.835 ± 0.028, demonstrating strong discriminatory performance for predicting progression to MM. This represents an 11% increase in predictive potential when compared to baseline risk assessment using the Mayo 2/20/20 criteria alone in the same cohort. Visualization of the model’s learned embeddings via UMAP confirmed that it successfully separated progressor and non-progressor trajectories into distinct clusters over time. The resulting patient dashboards provide a robust and interpretable tool that visualizes how changes in a patient’s biomarkers directly translate into an updated risk score. Conclusions: CALM leverages routinely collected serial laboratory data to dynamically update individualized risk predictions for SMM patients. This ‘digital twin’ approach not only provides improved risk stratification over static models but also offers clinicians interpretable feedback on which laboratory trends are driving risk, potentially informing more personalized monitoring and therapeutic decision-making in clinical practice.