This paper describes the development of an easy-to-use know- ledge-base platform to map perceptual features from sensory data to low-dimensional feature space. Our system enables the composition of perceived events to trigger salient features that are deemed to be of special interest to the robot. In generating the salient features, we factored in the robot’s internal states such as personality, characteristics, and more. In previous work, we employed a participatory design study to elicit the possible low-dimensional salient features and the corresponding trigger/s befitting a social robot. Here, we extract the technical requirements from that elicitation study and develop a knowledge-centric system design. Moreover, we develop a framework for non-experts to easily compose salient features from the multiple firings of high-dimensional signals even without the need for programming knowledge. This allows experts in various fields to easily design the robot’s reactive behaviors. Lastly, we evaluate our system’s response using data from a real human-robot interaction setup and verify its performance.