Abstract: Automatic failure diagnosis and correction may be performed on trained machine learning models. Input data that causes a trained machine learning model may be identified in order to determine different model failures. The model failures may be clustered in order to determine failure scenarios for the trained machine learning model. Examples of the failure scenarios may be generated and truth labels for the example scenarios obtained. The examples and truth labels may then be used to retrain the machine learning model to generate a corrected version of the machine learning model.

Abstract: Systems and methods are provided for controlling interactions between customers and agents. One embodiment is a system that includes an interface which receives a request to initiate a conversation. The system also includes a controller that acts as a message broker for the conversation by initiating a message stream between a customer device and an agent device, and records a time from receipt of each customer message to receipt of a corresponding agent response during the message stream. The controller also identifies an issue that the conversation is directed to, adjusts a permitted time period for agent responses based on a complexity of the issue, calculates an effectiveness score for the agent based on a number of agent responses that were provided within the permitted time period, updates a profile for the agent with the effectiveness score, and uses the effectiveness score as criteria for selecting the agent.