In a wireless system, the frequency-reconfigurable RF components are highly desired because one such component can replace multiple RF components to reduce the size, cost, and weight. Typically, the reconfigurable RF components are realized using capacitive varactors, PIN diodes, or MEMS switches. Most of these RF switches are expensive, rigid, and need tedious soldering steps, which are not suitable for futuristic flexible and wearable applications. Therefore, there is a need to have a solution for low cost, flexible, and easy to integrate RF switches. All the above-mentioned issues can be alleviated if these switches can be simply printed at the place of interest. In this work, we have demonstrated vanadium dioxide (VO2) based RF switches that have been realized through additive manufacturing technologies (inkjet printing and screen printing), which dramatically brings the cost down to a few cents. Also, no soldering or additional attachment step is required as the switch can be simply printed on the RF component. The printed VO2 switches are configured in two types (shunt configuration and series configuration) where both types have been characterized with two activation mechanisms (thermal activation and electrical activation) up to 40 GHz. The measured insertion loss of 1-3 dB, isolation of 20-30 dB, and switching speed of 400 ns are comparable to other non-printed and expensive RF switches. As an application for the printed VO2 switches, a fully printed frequency reconfigurable filter has also been designed in this work. An open-ended dual-mode resonator with meandered loadings has been co-designed with the VO2 switches, resulting in a compact filter with decent insertion loss of 2.6 dB at both switchable frequency bands (4 GHz and 3.75 GHz). Moreover, the filter is flexible and highly immune to the bending effect, which is essential for wearable applications. Finally, a multi-parameter (switch thickness, width, length, temperature) model has been established using a customized artificial neural network (ANN) to achieve a faster simulation speed. The optimized model’s average error and correlation coefficient are only 0.0003 and 0.9905, respectively, which both indicate the model’s high accuracy.