The aggregation of Distributed Energy Resources (DERs) into Virtual Power Plants (VPPs) is a critical strategy for managing the volatility of renewable energy. The effectiveness of a VPP hinges on a sophisticated software stack capable of forecasting, optimization, and real-time control. However, the development and benchmarking of these underlying machine learning (ML) models are often hindered by a lack of integrated, open-source research platforms. To bridge this gap, we present VPP-Sim, a modular, containerized simulation framework designed to accelerate the research and development lifecycle for VPP control strategies. VPP-Sim provides an end-to-end MLOps-ready pipeline, encompassing synthetic data generation, real-time streaming via Kafka, a model-agnostic forecasting engine with MLflow, and a linear programming optimization core. In this paper, we detail its microservices-based architecture, present the formalization of its economic dispatch optimization problem, and showcase its user interface and generated data profiles. The entire framework, including all code, data generation scripts, and deployment instructions, is made available open-source. We conclude by discussing a wide range of potential applications, including predictive maintenance and asset management, and outline a clear roadmap for future extensions. VPP-Sim is positioned as a foundational tool to empower researchers in the development of robust, reliable, and transparent ML solutions for sustainable power systems.