receiving, by a feature generation component, first data representing a feature definition for a machine learning model of a skill ranking component;

receiving first input data representing a user request;

determining, from the first input data, natural language understanding (NLU) data representing the user request, the NLU data comprising skill identifier data and intent data representing a semantic interpretation of the user request;

generating, by the feature generation component, first feature data from the skill identifier data and the intent data using the feature definition;

sending, by the feature generation component, the first feature data to the skill ranking component;

storing, by the skill ranking component, the first feature data in a first cache local to the skill ranking component;

storing, by the feature generation component, the first feature data in a second cache local to the feature generation component; and

generating, by the skill ranking component, a ranked list of speech processing skills for processing the user request by inputting the first feature data into the machine learning model of the skill ranking component.