Wastewater treatment operations can be significantly impacted by wet weather conditions, stressing frontline staff as well as the collection systems and treatment facilities they manage to their peak capacities. While managers have relied on weather forecasts to anticipate these conditions, they have lacked tools to translate these forecasts into precise, local, and actionable predictions of flows within their facilities. This presentation introduces a novel approach that integrates an ensemble of hydrologic prediction algorithms that combine machine learning (ML) and process-based rainfall-runoff models, with live weather observations and forecasts. This system operates in real time, delivering a continually updated dashboard of flow predictions to support critical operational decisions. Predicting impending wet weather events and attendant inflows to treatment facilities is vital for efficient management of collection and treatment operations. Historically, operators have relied on weather forecasts and mental models to predict future outcomes. Coupling the weather forecasts with predictive hydrological models enables more proactive planning to manage operations, reduce overflows, and flooding. Operators can also better prepare in advance, mobilizing treatment resources and protecting treatment performance in highly stressed conditions. Process-based hydrologic models have existed for decades having been applied in the design of many built assets. Historically, they have not been used for real-time applications necessary to support operations as the best models are computationally expensive, even with modern computing capabilities. Process-based models have an advantage, when properly calibrated, to predict outcomes for a range of events extending beyond previously observed data. More recently, data-driven or machine learning (ML) models are being used as alternatives to traditional hydrological predictions because of their computational efficiency and relatively accurate forecasts within the boundaries of training data. Because these models inform critical decisions and must handle higher-than-recorded flows seen today and anticipated in the future, developing more reliable and robust models is needed. This study presents an inflow prediction application for the Amherst, NY wastewater treatment plant, where ML approaches are integrated with physics-based principles that explain rainfall-runoff and flow routing mechanisms. Three advanced, data-driven variational deep learning models—non-linear autoregressive inputs (NARX), long short-term memory networks (LSTM), and transformer architectures—are compared to hybrid models aligned with rainfall-runoff and infiltration models, specifically the rainfall time kernel (RTK) unit hydrograph method and the antecedent moisture model (AMM). All models demonstrated strong performance when compared to observed data. However, those that incorporated process-based components showed higher accuracy and robustness, particularly when predicting peak flows during cross-validation. This study details how these ensemble models function as digital twins, enhancing situational awareness and providing decision support for city operators. By integrating machine learning-based into operational practices, municipalities can shift from reactive to proactive and resilient collection system management, developing targeted strategies and ensuring compliance with environmental regulations. As this field progresses, selecting appropriate modeling frameworks for different systems is crucial to ensure accuracy and reliability to support the increasingly critical decisions they facilitate in collection system management.