Distributed Internet-of-Things (IoT) environments, such as smart buildings, smart warehouses, and smart factory production lines are equipped with hundreds to thousands of sensors which produce streams of data. When aggregated, these data form a multi-variate time-series that frequently contain complex patterns that include seasonality, cycles, and varying amounts and type of noise. Building accurate predictive machine learning models for these environments is challenging. With the passage of time the statistical characteristics of the sensor data streams often change, causing concept drift that further degrades model performance. While there have been many efforts aimed at developing predictive models and new machine learning algorithms for IoT environments, there are few tools that allow researchers to explore the performance characteristics of their models and algorithms under highly realistic but tightly controlled conditions. In this paper we present the ASRL Mimic – a synthetic data generation framework that supports generation of simulated sensor stream data. With this library the researchers can control the seasonality and cycles present in the data, introduce trends, control the amount and type of noise, and simulate concept drift. They can use these capabilities to experimentally evaluate the performance of their models and algorithms. The ASRL Mimic can also analyze real IoT sensor data streams and generate very similar synthetic data.