Deceptive attacks exploiting the innate and the acquired vulnerabilities of human users have posed severe threats to information and infrastructure security. This work proposes INADVERT, a systematic solution that generates interactive visual aids in real-time to prevent users from inadvertence and counter visual-deception attacks. Based on the eye-tracking outcomes and proper data compression, the INADVERT platform automatically adapts the visual aids to the user's varying attention status captured by the gaze location and duration. We extract system-level metrics to evaluate the user's average attention level and characterize the magnitude and frequency of the user's mind-wandering behaviors. These metrics contribute to an adaptive enhancement of the user's attention through reinforcement learning. To determine the optimal hyper-parameters in the attention enhancement mechanism, we develop an algorithm based on Bayesian optimization to efficiently update the design of the INADVERT platform and maximize the accuracy of the users' phishing recognition.