Interactive assistant

An interactive troubleshooting assistant and method for troubleshooting a system in real time to repair (fix) one or more problems in a system is disclosed. The interactive troubleshooting assistant and method may include receiving multimodal inputs from sensors, wearable devices, a person, etc. that may be input into a feature extractor including attention layers and pre-processing units of a cloud computing system hosted by one or more servers, such as a private cloud system. A pre-processing unit converts the raw multimodal input into a structed form so that an attention layer can give weights to features provided by the pre-processing unit according to their importance. The weighted extracted features may be provided to an actions predictor. The actions predictor generates the most suitable action based on the weighted extracted features generated by the feature extractor based on the multimodal inputs. After the most suitable action is performed, the interactive troubleshooting assistant considers new information from multimodal inputs so that the interactive troubleshooting assistant can provide the next recommended action. The interactive troubleshooting assistant may repeat these operations until the repair is completed.