When engineers at Tonal thought about the challenge of providing personalized and dynamic guidance for strength training, we realized that we needed more than canned audio and video that would rarely be relevant to what the user was doing. We also believe that a real person giving guidance and support is the best way to deliver guidance and motivation. Video was the obvious choice, but a set composition can’t react to a user in real-time and can’t be personalized to match the user’s journey. Personalization would be limited to selecting the best video we have available for the user. We’ve developed a unique content creation and delivery system that allows flexible and personalized video content. The video technology built into Tonal works more like a live video editor than a typical video player. This enables our cloud AI infrastructure to deliver a personalized audio/video experience to the user. During production, audio and video are broken up into clips and we attach a variety of metadata about the contents of the clips. We have the ability to programmatically change any audio or video we insert into a workout. When we build the AV presentation for the user we choose the most relevant clips to insert into the workout. We can vary when and how much the coach appears. We can create progressions of form cues over the workout or even over the span of weeks or months based on the user journey. We take into account what we know about the user like their goals and previous workouts to adjust the curriculum and the audio and video experience. We can insert clips specific to the user’s progress or goals in any workout when that information is most relevant. The system also supports dynamically inserting audio feedback based on any number of metrics we are collecting about the user in real-time. We use this for simple applications like calling out the number of reps left in a set, and also more complicated metrics like form correction. As Tonal is able to detect more and more, the system allows improving existing content via additional audio that can be applied to all the content in the library. The talk would cover the details of this process. Presented at Demuxed 2020.