Protecting your home is not longer only the job of the police or a good dog. In the new social era we are living, we need smarter solutions, that can recognize and identify easily our friends, from intruders or unwanted presence. Such solutions should adapt to the home specifics, such as homes with noisy environments, random schedules, large variety of sounds, variety in the types of visual visitors, in order to avoid false alarms and keep your home secure. We are analyzing both audio (dog barking, glass breaking, steps, speech, alarm, etc.) and video analysis (humans, dogs, cats, faces, fire, etc.) in a parallel manner, to predict and classify events happening at home, and learn patterns from them in time. All machine learning algorithms are running on a small hardware device, which implies high optimization challenges for the performance of the algorithms using the low computation resources available. We are proud to be among the first ones to have such a large deployment of machine learning on a small custom-made IoT hardware device.