Approaches to dynamic medical playlist generation using polymorphic type data structures and systems for implementing the same

Abstract

Introduced here are computer-implemented platforms (also referred to as "playlist generation platforms") that are enabled to generate personalized activity sequences dynamically based on obtained user states. Machine learning model(s) are applied to user state data to identify activity domains that correspond to the user's current context and needs. A series of candidate activities are generated based on the identified domains using corresponding domain-specific models. The candidate activities are assembled into polymorphic data structures that accommodate multiple activity types, and personalized playlists are generated from the assembled activities. In some implementations, updated activity sequences are presented on an interface based on the evolving state of the user during activity execution. In response to receiving user feedback, a playlist generation platform adjusts the domain-specific models based on the user feedback.

Description

Attorney Docket No. 125847.8045. WO01APPROACHES TO DYNAMIC MEDICAL PLAYLIST GENERATION USING POLYMORPHIC TYPE DATASTRUCTURES AND SYSTEMS FOR IMPLEMENTING THE SAMECROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Application No.: 63 / 684,277, titled “Dynamic Medical Playlist Generator with Polymorphic Types Backed by Holistic User State-Based Recommendation Engines” and filed August 16, 2024, the entirety of which is incorporated herein by reference.TECHNICAL FIELD

[0002] Various embodiments concern playlist generation systems, and more particularly computer programs and associated computer-implemented techniques for providing personalized activity sequences for medical applications using machine learning models.BACKGROUND

[0003] A playlist (also referred to as a “compilation,” “collection,” “list,” and so forth) refers to a structured sequence of activities that are organized and presented to a user for completion (or presentation) within a defined timeframe. For example, playlists may contain activities of varying types, durations, and complexity levels, and can be used to achieve specific therapeutic or wellness objectives. An activity (also referred to as a “task” or “action”) represents a discrete task or action that a user is enabled to perform or be presented with through a digital interface, such as completing an exercise routine, answering health assessment questions, viewing educational content, or logging physiological data. Activities may be characterized by different data structures, execution requirements, and user interaction modalities.1183067633.1Attorney Docket No. 125847.8045. WO01BRIEF DESCRIPTION OF THE DRAWINGS

[0004] Figure 1 illustrates an example environment that includes a playlist generation platform used to generate playlists using machine learning (ML).

[0005] Figure 2 illustrates a network environment that includes a playlist generation platform that is executed by a computing device.

[0006] Figure 3 illustrates an example of a computing device enabled to execute a playlist generation platform.

[0007] Figure 4 illustrates an example environment of an architecture of a playlist generation platform used to generate personalized playlists for a user.

[0008] Figures 5A-5C illustrate example environments of an orchestrator engine within the playlist generation platform that manages multiple domain services.

[0009] Figure 6 is a flow diagram illustrating an example process of generating a personalized playlist for a user using a playlist generation platform.

[0010] Figure 7 illustrates an example environment of a modular front-end architecture used by the playlist generation platform.

[0011] Figures 8A and 8B illustrate a user interface presented to a user (e.g., a patient, a care provider) that displays an indication of a generated playlist.

[0012] Figure 9 illustrates a user interface displays an indication of a sequence of activities within a generated playlist.

[0013] Figure 10 illustrates a user interface that displays an indication of a particular activity (here, an exercise) within a generated playlist.

[0014] Figure 11 illustrates a user interface that displays pausing a particular activity (here, an exercise) within a generated playlist.

[0015] Figure 12 illustrates a series of screens of a user interface throughout a progression of a patient interacting with a generated playlist including activities belonging to a video modality.

[0016] Figure 13 illustrates a series of screens of a user interface throughout a progression of a patient interacting with a generated playlist including2183067633.1Attorney Docket No. 125847.8045. WO01 activities belonging to a computer vision modality.

[0017] Figure 14 illustrates a user interface that enables user interaction to receive user feedback.

[0018] Figure 15 illustrates a user interface that enables user interaction for a logging activity (here, a health logging activity) within the generated playlist.

[0019] Figure 16 illustrates a user interface that enables user interaction for an education activity within the generated playlist.

[0020] Figure 17 illustrates a layered architecture of an Al system that can implement the ML models of a playlist generation platform, in accordance with some implementations of the present technology.

[0021] Figure 18 is a block diagram showing some of the components typically incorporated in at least some of the computer systems and other devices on which a playlist generation platform operates in accordance with some implementations of the technology.

[0022] Various features of the technology described herein will become more apparent to those skilled in the art from a study of the Detailed Description in conjunction with the drawings. Various embodiments are depicted in the drawings for the purpose of illustration. However, those skilled in the art will recognize that alternative embodiments may be employed without departing from the principles of the technology. Accordingly, although specific embodiments are shown in the drawings, the technology is amenable to various modifications.3183067633.1Attorney Docket No. 125847.8045. WO01DETAILED DESCRIPTION

[0023] Exercise therapy is a clinical intervention technique that uses structured physical activities as the primary treatment modality for addressing symptoms of musculoskeletal conditions, chronic pain disorders, and various other health conditions. Exercise therapy programs generally involve systematic plans for performing specific physical activities during scheduled sessions that occur on a regular basis. The primary objectives of exercise therapy programs are to restore normal physical functionality, reduce pain levels, improve mobility, and enhance overall quality of life for individuals suffering from acute or chronic physical ailments. These programs are typically designed by healthcare professionals and tailored to address specific conditions, patient capabilities, and therapeutic goals. Exercise therapy activities may include guided movement routines, strength-building exercises, flexibility training, balance improvement tasks, and functional movement assessments that are specifically selected to address individual patient needs and treatment objectives.

[0024] In the context of digital health platforms, exercise therapy activities represent discrete, executable tasks that users perform as part of their therapeutic regimen, but they constitute only one type of activity within a broader ecosystem of therapeutic interventions. When multiple activities are organized into a structured sequence, they form what is commonly referred to as a “playlist” or “session.” These playlists may combine exercise therapy activities with other therapeutic modalities such as educational content, mindfulness exercises, health assessments, pain tracking, and behavioral interventions. The playlist format allows for logical progression through different activity types, thus enabling users to receive a comprehensive therapeutic experience that addresses multiple aspects of their condition, which often requires addressing not just physical symptoms but also educational needs, psychological factors, and behavioral patterns that contribute to overall health and wellness.

[0025] However, traditional digital health platforms face significant challenges in delivering personalized therapeutic content that adapts to users'4183067633.1Attorney Docket No. 125847.8045. WO01 evolving needs and circumstances. Most existing systems rely on static, preprogrammed content (e.g., activity) sequences that fail to account for changes in user state, such as fluctuations in pain levels, variations in physical capability, changes in environmental conditions, or shifts in motivation and engagement. The static approach results in therapeutic programs that may become irrelevant or inappropriate as users progress through their treatment or experience day- to-day variations in their condition. Furthermore, conventional systems typically focus on single activity types (e.g., modalities) per playlist, such as exercise routines alone, and generate new playlists for other complementary therapeutic modalities such as education or health tracking activities that improve overall treatment effectiveness.

[0026] Further, traditional digital health platforms often rely on limited data sets, such as basic demographic information, historical activity completion rates, or simple preference indicators to categorize the user into a particular user state (also referred to as a user category). Traditional digital health platforms fail to use user state information, including physiological data, contextual factors such as time of day or location, device capabilities, current pain levels, stress indicators, sleep quality, and other environmental or personal factors when generating playlists for the user. For example, a traditional digital health platform generates playlists for large user populations by segmenting users into broad categories such as “chronic back pain patients,” “post-surgical recovery patients,” or “general wellness users” based solely on their initial health assessment or diagnosis. All users within each category receive identical playlist sequences, such as a standardized “Week 1 Back Pain Program,” regardless of whether a particular user is experiencing a pain flare.

[0027] The technical architecture of traditional digital health platforms presents additional challenges related to scalability and maintainability of personalized content delivery. Many current systems use monolithic recommendation engines that are difficult to update, extend, or customize for different therapeutic domains. When new activity types need to be added or existing recommendation logic needs to be modified, traditional digital health platforms often require extensive redevelopment and / or testing. Further, many traditional digital health platforms use separate systems or applications for5183067633.1Attorney Docket No. 125847.8045. WO01 different activity modalities. The fragmented approach leads to increased downtime when updating the recommendation logic and duplicated development efforts.

[0028] Introduced here is an approach to providing personalized activity recommendations dynamically based on holistic user states. The approach not only solves the problem of delivering relevant therapeutic content but can also provide adaptive playlist generation without manual curation by healthcare professionals (e.g., physiotherapist, nurse, or physician) for each individual user. Simply put, the approach allows individuals to receive high-quality, personalized therapeutic programs at scale. As further discussed below, the approach may rely on previously obtained and / or real-time or near-real time user state data that is collected for an individual (e.g., from completing activities, providing feedback, engaging with content). This user state data is used by ML models trained on such data to generate activities that are presented for display on an interface that is accessible via a computing device. Generally, the computing device is associated with the individual and causes execution of the activities and collecting the interaction data from which the recommendations are generated.

[0029] The approach is implemented by a system (hereinafter the “playlist generation platform”) that is enabled to provide targeted activity sequences through polymorphic data structures (e.g., playlists containing different activity types such as exercise therapy sessions, educational content, health assessments, and pain tracking activities). As used herein, a “polymorphic data structures” refer to unified data frameworks that can accommodate different content types, such as activities with different execution requirements, user interface modalities, and data collection mechanisms, within a single organizational structure. ML model(s) are applied to user state data to identify activity domains that correspond to the user's current context and needs. User state data may include, for example, physiological measurements such as pain levels or heart rate variability, contextual information such as time of day or geographic location, device capabilities such as screen size or available sensors, historical interaction patterns such as completion rates or engagement6183067633.1Attorney Docket No. 125847.8045. WO01 metrics, and environmental factors such as weather conditions or ambient noise levels.

[0030] A series of candidate activities are generated based on the identified domains using corresponding domain-specific trained on training data of the corresponding domain. For example, an exercise therapy domain model generates candidate activities such as specific strength-building routines, while an education domain model generates or otherwise identifies articles, instructional videos, or interactive learning modules. The candidate activities are assembled into a polymorphic data structure, and a personalized playlist is generated from the assembled activities. In some implementations, updated activity sequences are presented on an interface based on the evolving state of the user during activity execution, such as adjusting subsequent activities in real-time if a user reports increased pain during an exercise. In response to receiving user feedback (e.g., difficulty ratings, pain assessments, completion status, or explicit preference indicators), the playlist generation platform adjusts parameters of the domain-specific models based on the user feedback.

[0031] The polymorphic data structure enables the combination of different activity types within a single playlist data structure. Additionally, the modular architecture enables simple expansion and customization (e.g., adding new activity types or modifying recommendation logic) without requiring extensive system redesign. Further, the playlist generation platform’s implementation of a continuous feedback loop enables the continuous improvement of generated playlists by refining the domain-specific models (i.e., the models that generate the activities in the playlist) over time. By continuously monitoring user interactions, physiological responses, and contextual changes, the playlist generation platform can adapt recommendations (e.g., in near-real-time or in real-time), ensuring that users receive up-to-date therapeutic content for their current state.

[0032] For the purpose of illustration, embodiments may be described with reference to exercise therapy activities that are performed during playlists as part of a therapeutic program. However, the playlist generation platform could be designed to generate playlists containing other activity types, such as7183067633.1Attorney Docket No. 125847.8045. WO01 entertainment media, productivity tasks, social interaction activities, creative exercises, skill-building modules, and the like. Accordingly, the approach described herein could be used to provide personalized playlist generation for nearly any combination of activities.

[0033] Moreover, embodiments may be described in the context of computer-executable instructions for the purpose of illustration. However, aspects of the approach could be implemented via hardware or firmware instead of, or in addition to, software. As an example, the playlist generation platform may be embodied as a computer program that offers support for completing activities during personalized sessions as part of a therapeutic program, determines which activity combinations are appropriate for a user given their holistic state data and feedback from past sessions, and enables communication between the user and one or more care providers (also referred to as “healthcare professionals,” “coaches,” or “clinicians”). The term “care provider” may be used to generally refer to individuals who monitor, guide, or otherwise facilitate engagement by users with the playlist generation platform. Care providers may include healthcare professionals such as physical therapists, nurses, or physicians, but could also include coaches, peer support, or other facilitators in some embodiments.Terminology

[0034] References in the present disclosure to “an embodiment” or “some embodiments” mean that the feature, function, structure, or characteristic being described is included in at least one embodiment. Occurrences of such phrases do not necessarily refer to the same embodiment, nor are they necessarily referring to alternative embodiments that are mutually exclusive of one another.

[0035] Unless the context clearly requires otherwise, the terms “comprise,” “comprising,” and “comprised of” are to be construed in an inclusive sense rather than an exclusive or exhaustive sense. That is, in the sense of “including but not limited to.” The term “based on” is also to be construed in an inclusive sense. Thus, the term “based on” is intended to mean “based at least in part on.”8183067633.1Attorney Docket No. 125847.8045. WO01

[0036] The terms “connected,” “coupled,” and variants thereof are intended to include any connection or coupling between two or more elements, either direct or indirect. The connection or coupling can be physical, logical, or a combination thereof. For example, elements may be electrically or communicatively coupled to one another despite not sharing a physical connection.

[0037] The term “module” may refer broadly to software, firmware, hardware, or combinations thereof. Modules are typically functional components that generate one or more outputs based on one or more inputs. A computer program may include or utilize one or more modules. For example, a computer program may utilize multiple modules that are responsible for completing different tasks, or a computer program may utilize a single module that is responsible for completing all tasks.

[0038] When used in reference to a list of multiple items, the word “or” is intended to cover all of the following interpretations: any of the items in the list, all of the items in the list, and any combination of items in the list.Overview of Playlist Generation Platform

[0039] A playlist generation platform may be responsible for creating and delivering personalized activity sequences for individuals (also called “users,” “patients,” or “participants”) through analysis of user state data that contains their health and / or other contextual information. As an example, the playlist generation platform may process user data through orchestration operations (such as through an orchestration engine) that are performed as part of the playlist generation process (or simply “process”) by identifying relevant activity domains and assembling activities received from corresponding domainspecific models of the relevant activity domains into a playlist. The platform may be required to generate playlists on a periodic basis. The frequency with which the platform generates playlists may be based on factors such as the type of therapeutic program, the user's condition severity, the availability of new activities, the complexity of the user's needs, the level of personalization required, and the like. In some embodiments, the playlist generation is generally performed through a continual analysis of user state data.9183067633.1Attorney Docket No. 125847.8045. WO01

[0040] As the platform generates playlists, it may receive input from various data sources of a computing system. For example, the data input interface is part of the computing system on which the playlist generation platform is executed or accessed. In order to initiate the playlist generation process, the playlist generation platform may activate one or more computer-executable workflows configured on, and executable by, a server or cloud instance, and the computer-executable workflows may instruct the playlist generation platform to ingest user state data in such a manner that one of its interfaces can process the information as playlists are assembled. Note that, in some embodiments, the data input interface is part of another computing system. For example, the interface may be included in a peripheral computing device, such as a wearable sensor or mobile application, that is linked to the computing system. By examining the user state data that is received by the interface, the playlist generation platform can create personalized activity sequences by identifying relevant domains and selecting appropriate activities over time.

[0041] As mentioned above, the playlist generation platform can alternatively generate personalized content sequences in contexts that are unrelated to healthcare, for example, to provide customized learning experiences in educational settings, personalized entertainment recommendations, or adaptive productivity workflows. As an example, the playlist generation platform may process the preferences and behavioral patterns of individuals within educational platforms, entertainment systems, professional development programs, etc. Accordingly, while embodiments may be described in the context of a system that generates playlists during a therapeutic intervention, the features of those embodiments may be similarly applicable to generating other types of personalized content sequences. Organizations or individuals who receive personalized playlists may be referred to as “users” of the playlist generation platform, even if these organizations or individuals have limited direct interaction with the playlist generation platform.

[0042] Figure 1 illustrates an example environment 100 that includes a playlist generation platform used to generate playlists using ML. In some implementations, the example environment 100 is implemented by a system (e.g., the playlist generation platform) including components of the example10183067633.1Attorney Docket No. 125847.8045. WO01 processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0043] The example environment 100, as shown in Figure 1 , includes an application 102, device capabilities 104, a playlist generation platform 106, and an orchestrator engine 108. Application 102 (e.g., a native or web-based application) operates as the user-facing interface of the system. Application 102 communicates directly or indirectly with the playlist generation platform 106 and manages the input and output of user data, the presentation of activity playlists, and / or the reception of user-generated feedback. The application 102 represents a software module executed on a user’s terminal device, such as a mobile phone, tablet, or computer. Through application 102, users receive personalized playlists and interact with prescribed activities, submit data about their session participation, and provide subjective measures such as pain ratings or difficulty levels.

[0044] Device capabilities 104 refer to one or more operations performable by or functions available to the terminal device or related peripherals. Device capabilities 104 inform the playlist generation platform of the range of sensor modalities, interface characteristics, and computational features available in the device environment. Such capabilities include, but are not limited to, the presence or absence of accelerometers, gyroscopes, cameras, heart rate monitors, local memory size, processing capability, and display or audio output features. The playlist generation platform 106 receives device capabilities 104 to adapt the format, type, and modality of recommended activities for compatibility with specific device environments. In some embodiments, device capabilities 104 are determined by querying a device's hardware abstraction layer or by retrieving device profiles (e.g., ones established during application onboarding). In some embodiments, device capabilities 104 are dynamically updated through runtime detection as device features become accessible or restricted — for example, when an external wearable sensor is paired or disconnected. In some embodiments, the device capabilities 104 include1 1183067633.1Attorney Docket No. 125847.8045. WO01 additional data fields sourced from paired medical devices, environmental sensors, or external data repositories.

[0045] The playlist generation platform 106 generates personalized activity playlists using ML. Playlist generation platform 106 ingests user state data and device capabilities 104, interfaces with the orchestrator engine 108, invokes domain-specific models in accordance with information received from the orchestrator engine 108, and assembles activities into polymorphic data structures (i.e., the playlists). In some embodiments, the playlist generation platform 106 is implemented as a microservices architecture composed of distinct submodules for user profiling, domain identification, activity selection, playlist assembly, feedback interface, adaptive logic, and so forth. In other embodiments, playlist generation platform 106 is realized as a distributed system that uses edge and cloud computing infrastructure to support the operations.

[0046] In operation, the device capabilities 104 are packaged as structured metadata and transmitted during session initiation and / or continuously as part of the active session data stream to the playlist generation platform 106. The playlist generation platform 106 transmits the received device capabilities to the orchestrator engine 108. The orchestrator engine 108 manages data flow within playlist generation platform 106, such that user input, device capability data, and contextual factors trigger specific downstream domain-specific logic (e.g., domain-specific models). The orchestrator engine 108 invokes, in succession or in parallel, domain-specific models for activity generation and / or prioritization. Subsequently, the playlist generation platform 106 (e.g., via the orchestrator engine 108) assembles a polymorphic data structure that indexes candidate activities by type, domain, and device compatibility. The resulting playlist is transmitted back to the application 102, which presents the activities to the user. For example, the presentation is via an ordered interface with modal transitions, progress indicators, and so forth. The playlist generation platform 106 can enable various functions such as pausing, resuming, and / or reordering activities. Each time the user interacts with an activity or provides feedback, the application 102 propagates the information to the playlist generation platform 106, which is enabled to use the feedback to update parameters of the invoked12183067633.1Attorney Docket No. 125847.8045. WO01 domain-specific models.

[0047] Figure 2 illustrates a network environment 200 that includes a playlist generation platform 202 that is executed by a computing device 204. Users can interact with the playlist generation platform 202 via interfaces 206. For example, users may be able to access interfaces that are designed to input user state data, view generated playlists, or configure personalization parameters. As another example, users may be able to access interfaces through which activity recommendations and feedback can be reviewed, completed, or shared with care providers. Thus, interfaces 206 may serve as data input and configuration spaces, or the interfaces 206 may serve as therapeutic delivery and interaction spaces through which users and care providers can engage with the personalized playlist content.

[0048] As shown in Figure 2, the playlist generation platform 202 may reside in a network environment 200. Thus, the computing device on which the playlist generation platform 202 is executing may be connected to one or more networks 208A-B. Depending on its nature, the computing device 204 could be connected to a personal area network (PAN), local area network (LAN), wide area network (WAN), metropolitan area network (MAN), or cellular network. For example, if the computing device 204 is a mobile phone, then the computing device 204 may be connected to a server system 210 via the Internet. As another example, if the computing device 204 is a computer server, then the computing device 204 may be accessible to users via respective computing devices that are connected to the Internet via LANs.

[0049] The interfaces 206 may be accessible via a web browser, desktop application, mobile application, or another form of computer program. For example, to interact with the playlist generation platform 202, a user may initiate a mobile application on the computing device 204 and then access the playlist generation features. As another example, a user may access, via a desktop application or mobile application, interfaces that are generated by the playlist generation platform 202 through which they can input user state data, configure therapeutic preferences, review generated playlists and activity recommendations, and the like. Accordingly, interfaces generated by the13183067633.1Attorney Docket No. 125847.8045. WO01 playlist generation platform 202 may be accessible via various computing devices, including mobile phones, tablet computers, desktop computers, wearable electronic devices, virtual reality systems, and augmented reality systems.

[0050] Generally, the playlist generation platform 202 is hosted, at least partially, on the computing device 204 that processes the user state data to be analyzed for playlist generation, as further discussed below. For example, the playlist generation platform 202 may be embodied as a mobile application executing on a mobile phone or tablet computer. In such embodiments, the instructions that, when executed, implement the playlist generation platform 202 may reside largely or entirely on the mobile phone or tablet computer. Note, however, that the mobile application may be able to access a server system 210 on which other aspects of the playlist generation platform 202 are hosted.

[0051] In some embodiments, aspects of the playlist generation platform 202 are executed by a cloud computing service operated by, for example, Amazon Web Services®, Google Cloud Platform™, or Microsoft Azure®. Accordingly, the computing device 204 may be representative of a computer server that is part of a server system 210. Often, the server system 210 comprises multiple computer servers. These computer servers can include information regarding different therapeutic activities and domains; ML models for identifying relevant activity domains and generating personalized recommendations; orchestration logic for assembling polymorphic playlists; templates for generating activity sequences; user data such as holistic state information, activity history, and therapeutic goals; and other assets related to personalized healthcare.

[0052] Figure 3 illustrates an example of a computing device 300 that is able to execute a playlist generation platform 312. As mentioned above, the playlist generation platform 312 can facilitate the creation and delivery of personalized activity sequences, for example, by processing user state data and generating dynamic playlists. As shown in Figure 3, the computing device 300 can include a processor 302, memory 304, display mechanism 306, communication module 308, user state sensor 310A, audio output mechanism 322, and audio input14183067633.1Attorney Docket No. 125847.8045. WO01 mechanism 324. Each of these components is discussed in greater detail below.

[0053] Those skilled in the art will recognize that different combinations of these components may be present depending on the nature of the computing device 300. For example, if the computing device 300 is a computer server that is part of a server system (e.g., server system 210 of Figure 2), then the computing device 300 may not include the display mechanism 306, user state sensor 310A, audio output mechanism 322, or audio input mechanism 324, though the computing device 300 may be communicatively connectable to another computing device that does include a display mechanism, a user state sensor, an audio output mechanism, or an audio input mechanism.

[0054] The processor 302 can have generic characteristics similar to general-purpose processors, or the processor 302 may be an applicationspecific integrated circuit (ASIC) that provides control functions to the computing device 300. As shown in Figure 3, the processor 302 can be coupled to all components of the computing device 300, either directly or indirectly, for communication purposes.

[0055] The memory 304 may be comprised of any suitable type of storage medium, such as static random-access memory (SRAM), dynamic randomaccess memory (DRAM), electrically erasable programmable read-only memory (EEPROM), flash memory, or registers. In addition to storing instructions that can be executed by the processor 302, the memory 304 can also store data generated by the processor 302 (e.g., when executing the modules of the playlist generation platform 312) and produced, retrieved, or obtained by the other components of the computing device 300. For example, data received by the communication module 308 from a source external to the computing device 300 (e.g., user state sensor 310B) may be stored in the memory 304, or data produced by the user state sensor 310A may be stored in the memory 304. Note that the memory 304 is merely an abstract representation of a storage environment. The memory 304 could be comprised of actual integrated circuits (also referred to as “chips”).

[0056] The display mechanism 306 can be any mechanism that is operable15183067633.1Attorney Docket No. 125847.8045. WO01 to visually convey information to a user. For example, the display mechanism 306 may be a panel that includes light-emitting diodes (LEDs), organic LEDs, liquid crystal elements, or electrophoretic elements. In some embodiments, the display mechanism 306 is touch sensitive. Thus, a user may be able to provide input to the playlist generation platform 312 by interacting with the display mechanism 306. Alternatively, the user may be able to provide input to the playlist generation platform 312 through some other control mechanism.

[0057] The communication module 308 manages communications external to the computing device 300. For example, the communication module 308 may manage communications with other computing devices (e.g., server system 210 of Figure 2, or a wearable sensor peripheral). The communication module 308 may be wireless communication circuitry that is designed to establish communication channels with other computing devices. Examples of wireless communication circuitry include 2.4 gigahertz (GHz) and 5 GHz chipsets compatible with Institute of Electrical and Electronics Engineers (IEEE) 802.1 1 , also referred to as “Wi-Fi chipsets.” Alternatively, the communication module 308 may be representative of a chipset configured for Bluetooth®, Near Field Communication (NFC), and the like. Some computing devices, like mobile phones and tablet computers, are able to wirelessly communicate via separate channels. Accordingly, the communication module 308 may be one of multiple communication modules implemented in the computing device 300. As an example, the communication module 308 may initiate and then maintain one communication channel with a wearable sensor (e.g., via Bluetooth), and the communication module 308 may initiate and then maintain another communication channel with a server system (e.g., via the Internet).

[0058] The nature, number, and type of communication channels established by the computing device 300 - and more specifically, the communication module 308 - may depend on the sources from which data is received by the playlist generation platform 312 and the destinations to which data is transmitted by the playlist generation platform 312. Assume, for example, that the computing device 300 is representative of a mobile phone or tablet computer that is associated with (e.g., owned by) a user. In some embodiments the communication module 308 may only externally16183067633.1Attorney Docket No. 125847.8045. WO01 communicate with a computer server, while in other embodiments the communication module 308 may also externally communicate with a source from which to receive user state data. The source could be another computing device (e.g., a wearable device or sensor peripheral that includes a user state sensor 31 OB) to which the mobile device is communicatively connected. User state data could be received from the source even if the mobile phone generates its own user state data. Thus, user state data could be acquired from multiple sources, and these user state data may correspond to different aspects of the user's holistic condition. Regardless of the number of sources, user state data ,or analyses of the user state data, may be transmitted to the computer server for storage in a digital profile that is associated with the user. The same may be true if the playlist generation platform 312 only acquires user state data generated by the user state sensor 310A. The user state data may initially be analyzed by the playlist generation platform 312, and then the user state data, or analyses of the user state data, may be transmitted to the computer server for storage in the digital profile.

[0059] The user state sensor 310A may be any electronic sensor that is able to detect and convey information in order to generate user state data, generally in the form of physiological, contextual, or behavioral measurements. Examples of user state sensors include heart rate monitors, accelerometers, gyroscopes, GPS modules, ambient light sensors, and microphones. The user state sensor 310A may be part of a sensing module that is implemented in the computing device 300. In some embodiments, the user state sensor 310A is one of multiple user state sensors implemented in the computing device 300. For example, the user state sensor 310A could be included in a mobile phone or wearable device. Alternatively, the user state sensor 310A may be externally connected to the computing device 300 such that the user state sensor 310A captures user state data and sends the user state data to the playlist generation platform 312.

[0060] For convenience, the playlist generation platform 312 may be referred to as a computer program that resides in the memory 304. However, the playlist generation platform 312 could be comprised of hardware or firmware in addition to, or instead of, software. In accordance with embodiments17183067633.1Attorney Docket No. 125847.8045. WO01 described herein, the playlist generation platform 312 may include a processing module 314, domain routing module 316, orchestration module 318, and graphical user interface (GUI) module 320. These modules can be an integral part of the playlist generation platform 312. Alternatively, these modules can be logically separate from the playlist generation platform 312 but operate “alongside” it. Together, these modules may enable the playlist generation platform 312 to programmatically generate personalized activity sequences for users during therapeutic interventions, through analysis of user state data generated by the user state sensor 310A.

[0061] The processing module 314 can process user state data obtained from the user state sensor 310A over the course of a playlist generation operation (e.g., session, cycle, process). The user state data may be used to determine relevant activity domains as further discussed below. The user state data may be representative of a series of physiological measurements, contextual information, and behavioral indicators. These data may be captured by the user state sensor 310A over time, such that each data entry captures different aspects of the user's holistic condition. In some embodiments, these data may be representative of streams of real-time data that is captured by the user state sensor 310A. In such embodiments, the user state data could also be called “streaming data.”

[0062] The user state data may be used to identify relevant therapeutic domains as further discussed below. For example, the processing module 314 may perform operations (e.g., filtering noise, changing contrast, reducing size) to ensure that the data can be handled by the other modules of the playlist generation platform 312. As another example, the processing module 314 may temporally align the data with data obtained from another source (e.g., another user state sensor) if multiple data sources are to be used to establish the holistic user state of interest.

[0063] Moreover, the processing module 314 may process information input by users through interfaces generated by the GUI module 320. For example, the GUI module 320 may be configured to generate a series of interfaces that are presented in succession to a user as she completes activities as part of a18183067633.1Attorney Docket No. 125847.8045. WO01 playlist. On some or all of these interfaces, the user may be prompted to provide input. For example, the user may be requested to indicate (e.g., via a verbal command or tactile command provided via, for example, the display mechanism 306) that she is ready to proceed with the next activity, that she completed the last activity, that she would like to temporarily pause the playlist, etc. These inputs can be examined by the processing module 314 before information indicative of these inputs is forwarded to another module.

[0064] The domain routing module 316 may identify relevant activity domains for the user through analysis of user state data, in accordance with the approach further discussed below. Specifically, the domain routing module 316 can create, based on user state data (e.g., generated by the user state sensor 310A or user state sensor 31 OB), a mapping that specifies which activity domains are most appropriate for the user's current circumstances. For example, the domain routing module 316 can apply a computer-implemented model (or simply “model”) called a routing machine learning model to the user state data, so as to produce the domain mapping. In some embodiments the routing model is designed and trained to identify a predetermined number and / or type of activity domains (e.g., exercise therapy, education, or any combination thereof), while in other embodiments the routing model is designed and trained to identify all relevant activity domains that are appropriate given the user state data provided as input. The routing model could be a neural network that when applied to the user state data, analyzes the information to independently identify domains that are representative of each therapeutic need of interest.

[0065] The orchestration module 318 may generate personalized playlists based on the outputs produced by the domain routing module 316. Referring again to the aforementioned examples, the orchestration module 318 could assemble activity sequences and polymorphic data structures based on an analysis of the identified domains. Moreover, the orchestration module 318 may determine appropriate activity combinations and sequencing for the user based on the outputs produced by the domain routing module 316, in accordance with the approach further discussed below. Specifically, the orchestration module 318 may determine an appropriate playlist composition for the user based on19183067633.1Attorney Docket No. 125847.8045. WO01 their current holistic state, and a determination as to how their current needs compare to available activities within each identified domain.

[0066] Other modules could also be included in some embodiments. For example, the playlist generation platform 312 may include a training module (not shown) that trains the routing machine learning model that is employed by the domain routing module 316. As another example, the playlist generation platform 312 may include a feedback processing module (not shown) that analyzes user responses and updating the domain-specific models used by the orchestration module 318 to determine which activity combinations are appropriate for a given user state.

[0067] Similarly, other components could be implemented in, or accessible to, the computing device 300 in some embodiments. For example, some embodiments of the computing device 300 include an audio output mechanism 322 and / or an audio input mechanism 324. The audio output mechanism 322 may be any apparatus that is able to convert electrical impulses into sound. One example of an audio output mechanism is a loudspeaker (or simply “speaker”). Meanwhile, the audio input mechanism 324 may be any apparatus that is able to convert sound into electrical impulses. One example of an audio input mechanism is a microphone. Together, the audio output and input mechanisms 322, 324 may enable personalized recommendations and feedback to be audibly provided to the user during playlist execution. Assume, for example, that the user has been provided with a personalized playlist while being monitored by the user state sensor 310A. In such a scenario, the user may be audibly guided, in a personalized manner, via the audio output mechanism 322.Generating Playlists Using the Playlist Generation Platform

[0068] Various attempts have been made to improve the effectiveness of personalized playlist delivery while maintaining scalability and adaptability for diverse user populations. Consider, for example, a scenario where a healthcare provider needs to deliver personalized therapeutic interventions to thousands of users with varying conditions, preferences, and needs. The provider may struggle to create individualized treatment plans (i.e., playlists) due to resource20183067633.1Attorney Docket No. 125847.8045. WO01 constraints and the complexity of accounting for dynamic user states and contextual factors. Simply put, because manual curation of personalized content is time-intensive and static approaches fail to adapt to changing user circumstances, the provider may miss important opportunities to optimize therapeutic outcomes and user engagement.

[0069] Introduced here is an approach to providing Al-assisted playlist generation and personalization during therapeutic content delivery operations (e.g., during the assembly of exercise therapy, education, and wellness activities). With this approach, there are several advantages over conventional approaches that rely on static content sequences or simple rule-based recommendation systems. First, the playlist generation platform may use a ML system that can generate a personalized representation of targeted activity content from user state data. The playlist generation platform can identify relevant activity domains based on received user state data (even as the user state data changes). For example, the platform can identify and correlate environmental factors and device capabilities to determine appropriate activity types and sequencing. Second, the playlist generation platform can generate dynamic playlists based on the patterns and relationships derived from the user's state data and historical interactions.

[0070] One benefit of this advanced personalization approach is that the playlist generation platform can account for the limitations that have historically been introduced by static content delivery or simple categorization techniques. Traditionally, in order to provide therapeutic content at scale, healthcare providers were limited to working with broad user categories or one-size-fits-all program sequences. However, this restricted approach often led to missed therapeutic opportunities or suboptimal engagement due to misaligned content recommendations. As mentioned above, the playlist generation platform can apply ML models to user state data to generate personalized playlists. Due to the nature of its programming and training, the models can preserve subtle user preferences that may be lost in traditional static recommendation approaches.

[0071] Figure 4 illustrates an example environment 400 of an architecture of a playlist generation platform used to generate personalized playlists for a user.21183067633.1Attorney Docket No. 125847.8045. WO01In some implementations, the example environment 400 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0072] The application 402 operates as the user-facing interface for communication between users and the playlist generation platform. The application 402 operates as a digital interface, such as a mobile application, web application, or desktop client, that transmits user actions and parameters, retrieves personalized playlists, and presents activities in an ordered, interactive form. In some embodiments, application 402 is a modular application enabled to operate with multiple input and output modalities, including touch, voice, video, sensor data, and so forth.

[0073] The edge service 404 operates as an intermediary computational layer between the application 402 and a backend host of the playlist generation platform. The edge service 404 preprocesses, validates, and / or secures inbound and outbound payloads in proximity to the user device (i.e., a user device running the application 402). In some embodiments, the edge service 404 is deployed on edge gateways or access points distributed across a telecommunications network or, alternatively, as compute environments provisioned on-demand in cloud-infrastructure within each user’s geographic region.

[0074] The orchestrator engine 406 receives processed inputs from the edge service 404 and transmits the inputs (or transformed inputs) across multiple downstream modules to coordinate the assembly of personalized playlists. In some embodiments, the orchestrator engine 406 is realized as a service mesh controller that receives raw input, such as real-time sensor feeds, historical health records, session metadata, or external context signals, and applies transformation logic to map the input to user state representations. Further, the orchestrator engine 406 can use third-party data enrichment (e.g.,22183067633.1Attorney Docket No. 125847.8045. WO01 environmental context, social determinants, device logs) to provide additional covariates in mapping to the user state representation.

[0075] The activity domains 408 define a set of independent but interoperable clusters of therapeutic, educational, or behavioral content types that serve as categorical organizing structures for the activities populating a user playlist. The activity domains 408 can include different delivery modalities such as exercise therapy, education, assessment, or mindfulness. Each activity domain can be implemented as a microservice or a data repository housing domain-specific activity descriptions, logic, and / or ML models. In some embodiments, the activity domains 408 include additional programmatic interfaces that manage domain-specific rulesets, feedback ingestion pipelines, and so forth.

[0076] The activity generation engine 410 transforms outputs of the orchestrator engine 406 into activity sequences (i.e., the playlist). The activity generation engine 410 executes rule-based or machine learning-based decision logic to select, combine, and order activities from the routed domains into polymorphic playlist structures. The activity generation engine 410 can score, rank, prune, and sequence candidate activities depending on, for example, personalization targets, clinical efficacy, and engagement metrics. In some embodiments, the activity generation engine 410 uses reinforcement learning to refine the playlist based on received feedback.

[0077] The activity generation engine 410 can generate activities across different modalities, such as an education service, exercise service, and activity service. For example, an education service can be associated with educational modules, articles, and instructional videos. An exercise service can be associated with exercise routines, demonstrations, and movement assessments. An activity service can be associated with data schemas and interfaces for tracking, logging, and delivering generic activity types.

[0078] The user data 412 includes structured, semi-structured, and / or unstructured data associated with the individual user. The user data 412 can include static attributes (e.g., demographic, baseline risk factors, device capabilities) and / or dynamic state information (e.g., real-time physiological23183067633.1Attorney Docket No. 125847.8045. WO01 signals, behavioral patterns, engagement metrics, longitudinal outcome trajectories). The user data 412 can be continuously updated. The user context generation engine 414 synthesizes context representations from the user data 412 and / or other external context streams to generate descriptors of user state (e.g., indicators of activity readiness, adherence probability, and contextual risk factors) to inform the orchestration engine 406 and downstream domain selection or ranking. The user context generation engine 414 can apply rulebased, statistical, and / or ML-derived logic to normalize, interpolate, and abstract context variables used to map the user data 412 to particular user states.

[0079] For example, context can be generated by the user context generation engine 414 using a user service, which manages the persistence and integrity of user records, such as account management, authentication, authorization, and user-specific configuration. The user service ensures that only valid and authenticated sessions are granted access to playlist content and personal data. In some embodiments, user service exposes endpoints for querying, updating, or archiving user records. Further, user tagging can be used to tag and classify users along multiple axes or ontologies based on user characteristics, usage patterns, health trajectories, system interactions, and so forth. For example, the user context generation engine 414 can annotate user profiles with markers (e.g., indicators) for program eligibility, activity suitability, compliance risk, engagement propensity, and so forth. A member care data service can be used to integrate external data sources from care teams, EHR systems, or clinical coordination environments, to provide further user data 412. The member care data service can update records based on care outcomes, care team interactions, and / or remote monitoring data. The user context generation engine 414 can use a user meta service to track and manage supplementary metadata related to user interaction sessions, device usage, connectivity characteristics, content preference history, and environmental contextual data not otherwise captured by the user data 412. For example, the user meta service maintains stateful logs, session fingerprints, and auxiliary feature representations.

[0080] Figures 5A-5C illustrate example environments of an orchestrator24183067633.1Attorney Docket No. 125847.8045. WO01 engine within the playlist generation platform that manages multiple domain services. In some implementations, the example environment(s) are implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0081] Figure 5A illustrates an architecture for creating a playlist session to generate a playlist. A client 502 (e.g., a user-facing application or platformspecific session initiator) issues a session request (e.g., a request designated by a session UUID), which is transmitted to the orchestration layer 504. The orchestration layer 504 can forward the request payload to the orchestrator engine 506. The orchestrator engine 506 receives the session request, which can include information such as user identifiers, current user state (e.g., pain score, activity completion rates, device profile), from the orchestration layer 504.

[0082] Once the session request is received, the orchestrator engine 506 applies a routing machine learning model or other programmed routing logic trained to map the session request to a subset of relevant domain services (510a-510d). For example, the routing is determined by transforming incoming user state data (e.g., physiological sensor outputs, historical engagement metrics, contextual values such as device capabilities or environmental cues, and explicit user input) into a structured, high-dimensional feature vector. The orchestrator engine 506 can apply a rule-based engine or a machine learning classifier, which may be a multi-class or multi-label model (e.g., a neural network, ensemble method, or support vector classifier), that has been trained on historical user-state-to-domain mappings to determine the activity domains. In some embodiments, the orchestrator engine 506 uses a vector similarity algorithm, such as cosine similarity or Euclidean distance, between the realtime user state vector and a library of domain prototype vectors. The orchestrator engine 506 can apply a threshold or top-N selection criteria to the resulting scores, retaining only those domains whose similarity measure either25183067633.1Attorney Docket No. 125847.8045. WO01 exceeds a predefined threshold or ranks within the highest N domains.

[0083] The orchestrator engine 506 dispatches fetch activities calls, with session-specific parameters, to each domain service 510 (a first domain service 510a, a second domain service 510b, a third domain service 510c, a fourth domain service 51 Od, and so forth) associated with each routed activity domain. Each domain service 510 can apply its own domain-specific ML-based or programmed logic to generate one or more candidate activities. The orchestrator engine 506 receives the activity lists and can apply a set of assembly rules (e.g., activity type sequencing, duplication filters, global eligibility checks) to construct a polymorphic type data structure, such as an ordered or indexed list encoding the activities along with domain, metadata, and execution requirements. The orchestration layer 504 can transmit the assembled polymorphic playlist back to the client 502.

[0084] Figure 5B illustrates an architecture for tracking progress of the generated playlist. The orchestration layer 504 interfaces with the domain services 510 to coordinate the execution of discrete activities and / or the aggregation of playlist session data. As the user engages with the prescribed activities via a client 502, each activity’s completion state, performance metrics, and user-generated feedback (such as difficulty ratings, pain levels, or subjective assessments) can be captured during interaction sequences and transmitted from the client 502 to the orchestration engine 506. The orchestration engine 506 persists the activity engagement events through an event-driven architecture, such that a transition (e.g., completion, skipping, or pausing) is propagated to the respective domain services 510.

[0085] Each domain service can update the execution state of its respective activities and thereby maintain an up-to-date, session-scoped record of user actions, completion timestamps, and / or metadata, such as interruptions or deviations from the prescribed sequence. The updates are transmitted by the domain layer 508 as change data capture (CDC) events to an infrastructure layer 516 incorporating an event bus 518, which manages the asynchronous delivery of progress and state information.

[0086] Figure 5C illustrates an architecture for rehydrating paused or26183067633.1Attorney Docket No. 125847.8045. WO01 stopped playlist sessions of the generated playlist. When a playlist session is paused or stopped, the playlist generation platform determines the current execution state, such as the user's position within the playlist, the status of each partially or fully completed activity, associated feedback entries, and any temporal context such as timestamps or interruption causes. The serialized session state is persisted in a distributed data store accessible to both the orchestration engine 506 and the routed domain services.

[0087] Upon a user’s subsequent re-entry or explicit resume request, the orchestration engine 506 queries the stored session state and reconstructs the polymorphic playlist data structure (e.g., an array or list with activity domain references, prior completion flags, and user-provided metadata) representing the session point at interruption. The orchestration engine 506 routes this reconstructed session back to the orchestration layer 504 and client 502, enabling the user to continue interacting with the playlist.

[0088] Figure 6 is a flow diagram illustrating an example process of generating a personalized playlist for a user using a playlist generation platform. In some implementations, the example process 600 is performed by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0089] In operation 602, the playlist generation platform receives, via a network interface coupled to the computing device, user state data / information (e.g., physiological sensor data, user-reported survey responses, historical activity completion data, or real-time location information) of a user from a plurality of data sources (e.g., manually input data, stored data, sensor data).

[0090] In operation 604, the playlist generation platform accesses an activities repository that identifies a plurality of activities (or other items) for the user across multiple activity domains. Each activity corresponds to one or more activity domains. One or more activities of the plurality of activities are27183067633.1Attorney Docket No. 125847.8045. WO01 represented by different data types. In some embodiments, the activities repository includes metadata for each activity, such as a difficulty level, an estimated duration, an equipment list, or a target body region.

[0091] In operation 606, the playlist generation platform applies, to the user state data and the multiple activity domains, a routing machine learning model trained to map each portion of the user state data to a subset of activity domains within the multiple activity domains based on predefined information of the user and other portions of the user state data. The playlist generation platform can filter the plurality of items in the repository based on at least one user-specified exclusion criterion prior to determining the subset of item domains.

[0092] For each domain within the subset of activity domains, in operation 608, the playlist generation platform applies, to the user state data and respective activities of the activity domain, a respective domain-specific machine learning model associated with the activity domain that is trained to generate one or more candidate activities based on the user state data. Each domain-specific machine learning model can be trained on different training datasets. A training dataset can include labeled associations between user state data and activity domains. The playlist generation platform can train at least one domain-specific machine learning model using respective interaction data associated with multiple users.

[0093] In operation 610, the playlist generation platform generates a polymorphic type data structure by aggregating and indexing the one or more candidate activities for each domain across the subset of activity domains. In some embodiments, the polymorphic type data structure is an array or list in which each element includes a reference to a candidate activity and an identifier of the respective data type of the candidate activity. The playlist generation platform can prioritize the candidate items indexed within the polymorphic data structure based on a ranking score for each candidate item. For example, the ranking score is determined based on one or more of the user state information or historical user interaction data.

[0094] In operation 612, the playlist generation platform executes, via the network interface, the candidate activities indexed within the polymorphic type28183067633.1Attorney Docket No. 125847.8045. WO01 data structure by, for each candidate activity, transmitting computer-executable instructions to the computing device based on a respective data type of the candidate activity. For each candidate activity, the playlist generation platform can identify a user interface modality to present the activity based on the respective data type of the candidate activity. In some embodiments, the playlist generation platform generates a report that summarizes the data types and item domains of the candidate items over a particular time period.

[0095] In some embodiments, prior to executing the candidate activities, the playlist generation platform can filter the candidate activities based on a predefined set of user constraints. In some embodiments, subsequent to execution of at least one candidate activity, the playlist generation platform receives user feedback data indicating a response to the activity. The playlist generation platform can store the user feedback data in association with the corresponding activity and user state data. In response to feedback, the playlist generation platform can update at least one of the routing artificial intelligence model or a domain-specific artificial intelligence model based on received user feedback data.

[0096] For example, the playlist generation platform receives interaction data responsive to the transmitted executable instructions. The playlist generation platform can store the interaction data in a user profile database. The interaction data can be indexed by a particular item domain and a particular data type. The playlist generation platform can adjust one or more model parameters of one or more domain-specific artificial intelligence models based on the interaction data. The playlist generation platform generates a second polymorphic data structure by triggering the series of domain-specific artificial intelligence models to identify a second set of items within the repository based on a respective item domain of each item within the repository and the adjusted one or more parameters. For each item in the second polymorphic data structure, playlist generation platform causes transmission of a second set of executable instructions to the computing device, based on a respective data type of the identified second set of items. The playlist generation platform can log the adjustment to the one or more model parameters in a database. The playlist generation platform can generate a notification configured to be29183067633.1Attorney Docket No. 125847.8045. WO01 presented on the computing device that indicates that the second set of items is generated based on the interaction data. In some embodiments, the first set of executable instructions and the second set of executable instructions are each configured to trigger a different user interface presentation based on a respective data type of the respective item.

[0097] Figure 7 illustrates an example environment 700 of a modular frontend architecture used by the playlist generation platform. In some implementations, the example environment 700 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Those skilled in the art will recognize that different combinations of these components may be present depending on the nature of the playlist generation platform implementation. For example, if the environment 700 is a web-based application that is part of a larger therapeutic system, then the architecture may not include all specialized controllers or interface components, though the system may be communicatively connectable to other modules that do include these user experience elements.

[0098] The foundational modules 702 can interface with all components of the modular architecture, either directly or indirectly, for communication and resource sharing purposes. The foundational modules 702 can include foundational Al capabilities that provide machine learning functionality for user state analysis and activity recommendation, GraphQL interfaces that facilitate structured data queries across the platform, network request handlers that manage communication between client applications and backend services, exercise definition repositories that store and organize activity metadata and instructions, data cache systems such as Redis that store accessed user data and activity content, feature flags that enable the dynamic configuration and A / B testing of platform functionality, activity platform events that coordinate realtime updates and system notifications, mapped events that translate between different data formats and system interfaces, local device data management that store and / or synchronize user information on client devices, and so forth.30183067633.1Attorney Docket No. 125847.8045. WO01

[0099] The domain module wrapper 704 can facilitate communication between the foundational modules 702 and domain-specific functionalities of specialized controllers. In addition to managing interactions between foundational modules 702 and specialized controllers, the domain module wrapper 704 can also coordinate data flow generated by user interactions and produced, retrieved, or obtained by the other components of the modular architecture of the environment 700. For example, user state data received from external sensors can be transmitted through the domain module wrapper 704, and personalized recommendations produced by domain-specific models can be transmitted through the domain module wrapper 704. The domain module wrapper 704 can be an abstraction layer that includes software interfaces and API endpoints.

[0100] The setup controller 706 manages pre-session configuration via multi-phase control logic, shared business logic, shared eventing, and so forth. In some embodiments, the setup controller 706 manages user onboarding workflows. Thus, a user can configure playlist preferences and parameters of the playlist generation platform through interactions managed by the setup controller 706.

[0101] The overview screen 708 refers to a user interface component that presents a summary view of playlist content, therapeutic progress, or session information to users before, during, or after playlist execution. The overview screen 708 can display aggregated data (i.e., the overview content 710) such as completed activities, upcoming therapeutic content, or personalized recommendations in a consolidated format that enables users to quickly understand their current therapeutic status and planned activities.

[0102] The introductory screen 712 refers to a user interface component that presents initial information, instructions, or preparation content (i.e., the introductory content 714) to users at the beginning of a playlist session or individual activity. The introductory screen 712 can provide context (via instructional text, media, or interactive components) about upcoming therapeutic activities, explain how to interact with the content, or present motivational messaging to enhance user engagement and readiness for31183067633.1Attorney Docket No. 125847.8045. WO01 therapeutic intervention.

[0103] The session controller 716 manages active playlist sessions and coordinating user interactions. For example, the session controller 716 can coordinate activity sequencing and user experience flow. Examples of session management capabilities include playlist progression tracking, user state monitoring, activity transition management, updating playlists, and so forth. The nature, number, and type of session management capabilities can depend on the activity types included in generated playlists and the domains to which content is delivered by the playlist generation platform. Assume, for example, that the environment 700 is representative of a mobile application interface that is associated with (e.g., used by) a chronic pain management user. In some embodiments the session controller 716 only coordinates exercise therapy activities, while in other embodiments the session controller 716 also coordinates educational content, mindfulness exercises, and health assessments within playlist sessions. The activity types can be managed by specialized controllers (e.g., modality controller 718 or exercise controller 734) to which the session controller 716 is communicatively connected to. User interaction data can be processed through multiple controllers even if the session controller 716 manages overall playlist progression.

[0104] During the execution of the playlist (e.g., exercise therapy), The modality controller 718 manages different modalities (i.e., activity types) and coordinates modality-specific user experiences. Examples of modality management capabilities include managing state information, coordinating eventing, and implementing particular logic components depending on the modality. The modality controller 718 can be externally connected to the modular architecture such that the modality controller 718 coordinates with the session controller 716.

[0105] The immersive exploration controller 720 can manage immersive exercise functionality and coordinate user experiences over the course of playlist execution operations (e.g., session, activity sequence, therapeutic intervention). The immersive functionality can be used to increase user engagement with therapeutic content as further discussed below. The32183067633.1Attorney Docket No. 125847.8045. WO01 immersive capabilities can be representative of Al features, eventing, and business logic components. These capabilities can be coordinated by the immersive exploration controller 720 over time, such that each interaction captures different aspects of the user's engagement. In some embodiments, these capabilities are representative of user experience adaptations that are managed by the immersive exploration controller 720. In such embodiments, the immersive functionality could also be called “adaptive content delivery.”

[0106] The exercise controller 734 can be used to manage computer vision exercise functionality and coordinate submodular control as further discussed below. The exercise controller 734 can include functionalities used to guide users through exercise movements, coordinate pause and resume functionality during exercise sessions, handle error states, and synchronize exercise progress data with the broader playlist generation system.

[0107] After the execution of the playlist (e.g., exercise therapy), the post session controller 748 can manage information and experiences that occur after playlist completion through interfaces coordinated by various screen components. For example, the overview screen 708 and overview content 710 may present a series of interfaces that are displayed in succession to a user as she completes post-session activities. On some or all of these interfaces, the user may be prompted to provide feedback or review session outcomes. For example, the user may be requested to indicate (e.g., through interaction with introductory screen 712 or introductory content 714) that she completed the playlist successfully, that she experienced specific therapeutic benefits, that she would like to modify future playlist recommendations, etc. These interactions can be processed by the post session controller 748 before information indicative of these interactions is forwarded to other system components.

[0108] The immersive content 724 can present content and prepare users for the activity experiences through coordination with the interactive activity screen 726, in accordance with the approach further discussed below. Specifically, the immersive content 724 can create, based on playlist data and user preferences, immersive experiences that specify which activities will be33183067633.1Attorney Docket No. 125847.8045. WO01 presented and how users can interact with the content. For example, the immersive content 724 can coordinate with immersive activity content 728 to present therapeutic activities, so as to enhance user engagement. In some embodiments the immersive functionality is designed and configured to present a predetermined number and / or type of immersive experiences (e.g., guided exercise sessions, interactive educational content, gamified therapeutic activities, or any combination thereof), while in other embodiments the immersive system is designed and configured to present all relevant immersive content that is appropriate given the user's current playlist composition. The immersive coordination can include interactive puzzle scores 730 and immersive future context 732 that, when presented to users, evaluate engagement patterns to independently optimize or otherwise modify immersive experiences that are representative of each therapeutic activity type of interest.

[0109] The exercise preview screen 736 can present specific previews based on the outputs produced by activity context 738. Referring again to the aforementioned examples, the exercise preview screen 736 can coordinate exercise demonstrations and preparation experiences based on an analysis of the user's therapeutic needs and playlist composition. Moreover, the exercise activity 740 can manage active exercise experiences based on the outputs produced by activity future context 742, in accordance with the approach further discussed below. Specifically, the exercise activity 740 can coordinate appropriate exercise experiences for the user based on their current therapeutic state, and a determination as to how their current needs compare to available exercise content within the playlist.

[0110] Other interface components could also be included in some embodiments. For example, the modular architecture can include failure screen 744 and exercise future context 746 that coordinate error handling and progression management that are employed during activities. As another example, the modular architecture can include interactive 750 and incentives 752 that are manage gamification elements and motivation systems used by various controllers to determine which user experiences are appropriate for different therapeutic scenarios.34183067633.1Attorney Docket No. 125847.8045. WO01

[0111] Similarly, other interface components could be implemented in, or accessible to, the modular architecture in some embodiments. For example, some embodiments of the modular architecture include communication screen 754 and communication context 756, along with public overlay screen 758 and public overlay 760. The communication screen 754 can be any interface component that is able to facilitate user communication and care provider interaction. Examples of communication capabilities include messaging interfaces, progress sharing, care team coordination, or social support features. Meanwhile, the public overlay screen 758 and public overlay 760 can be any interface components that are able to manage shared experiences and community features to complement the personalized therapeutic content. Examples of public interface capabilities include peer support networks, community challenges, or shared progress tracking.

[0112] Figures 8A and 8B illustrate a user interface presented to a user (e.g., a patient, a care provider) that displays an indication of a generated playlist. In some implementations, the user interface(s) are implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0113] As shown in Figure 8A, the user interface presents a playlist activity preview 802 that displays personalized therapeutic content generated by the playlist generation platform. The playlist activity preview 802 can include visual representations of recommended activities, duration indicators, and contextual information that helps users understand the scope and nature of their personalized therapeutic session. The interface can also display a goal indicator 804 that tracks user progress toward therapeutic objectives over a specified time period, providing motivation and accountability through visual progress tracking. The goal indicator 804 can show completion ratios, target achievements, or milestone progress that is dynamically updated based on user engagement with generated playlists.35183067633.1Attorney Docket No. 125847.8045. WO01

[0114] Figure 8B illustrates a daily session interface 806 that presents users with a structured view of their personalized therapeutic activities for a given day. The daily session interface 806 can aggregate multiple activity types from different therapeutic domains into a single session experience, displaying the total estimated duration and providing users with an overview of upcoming activities. The interface can present activity categories such as exercise therapy, educational content, and health logging activities in a structured format. The user interfaces can include interactive elements such as start buttons, navigation controls, and progress indicators that enable users to initiate and manage their playlist sessions.

[0115] Figure 9 illustrates a user interface 900 displays an indication of a sequence of activities within a generated playlist. In some implementations, the user interface 900 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0116] The user interface presents an activity sequence 902 that displays a structured list of activities organized within a personalized playlist session. The activity sequence 902 can include multiple activity types such as introductory content, exercise therapy activities, and educational components, each with associated duration indicators and activity classifications. The sequence can present activities in a progression that is determined by the playlist generation platform (i.e., by the orchestrator engine, domain-specific models).

[0117] The interface can display an equipment indicator 904 that specifies any physical items or environmental requirements needed to complete the activities within the playlist. The equipment indicator 904 can provide users with notice of preparations to the environment prior to execution of the playlist, such as wall space for exercises, resistance bands, or other tools. The equipment information can be dynamically determined based on the specific activities36183067633.1Attorney Docket No. 125847.8045. WO01 selected by the domain-specific models and assembled into the polymorphic playlist structure. The activity sequence 902 can present each activity with visual thumbnails, descriptive titles, and so forth.

[0118] Figure 10 illustrates a user interface 1000 that displays an indication of a particular activity (here, an exercise) within a generated playlist. In some implementations, the user interface 1000 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0119] The user interface presents information about a specific activity selected from the polymorphic playlist structure. The interface can display an instructional video interface 1002 that provides users with visual demonstrations of proper exercise form, movement patterns, and techniques. The instructional video interface 1002 can include multimedia content (e.g., demonstration videos, animated guides, or step-by-step visual instructions) that helps users understand how to perform the activity effectively. The interface can present an activity modifications interface 1004 that offers alternative approaches or difficulty adjustments for the activity, enabling personalization based on user capabilities, pain levels, or equipment availability. Additionally, the interface can include an auto-start toggle 1006 that allows users to configure whether activities should automatically progress to the next item in the playlist sequence and provide control over session pacing.

[0120] Figure 1 1 illustrates a user interface 1 100 that displays pausing a particular activity (here, an exercise) within a generated playlist. In some implementations, the user interface 1100 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include37183067633.1Attorney Docket No. 125847.8045. WO01 different and / or additional operations or can perform the operations in different orders.

[0121] The user interface enables users to temporarily interrupt their activities. The interface can display a progress indicator that shows the user's current position within the overall playlist sequence, providing context about session completion status. When an activity is paused, the interface can present a pause menu 1 1102 that offers various options for session management, including the ability to restart the current activity, skip to the next activity, access instructional content (e.g., video demonstrations or written instructions), modify session settings, or terminate the entire session. The pause menu 11102 can include a resume function that allows users to continue from their current position, maintaining session state and progress tracking throughout the interruption.

[0122] Figure 12 illustrates a series of screens of a user interface 1200 throughout a progression of a patient interacting with a generated playlist including activities belonging to a video modality. In some implementations, the user interface 1200 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0123] The user interface presents overview screens 1202 that provide users with session summaries, activity previews, and preparation information before engaging with video content. The progression can include activity screens 1204 that display the actual video content (e.g., guided exercise demonstrations, educational presentations, or instruction videos) with integrated playback controls, progress tracking, and user interaction elements. The interface can show paired screens 1206 that present complementary information or interactive elements alongside the video content, such as exercise repetition counters, form feedback, or supplementary instructional38183067633.1Attorney Docket No. 125847.8045. WO01 text.

[0124] Figure 13 illustrates a series of screens of a user interface 1300 throughout a progression of a patient interacting with a generated playlist including activities belonging to a computer vision modality. In some implementations, the user interface 1300 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0125] As shown in Figure 13, the user interface demonstrates the user experience flow for computer vision-enabled activities that use near-real-time or real-time motion analysis and pose estimation. The interface can present tutorial screens 1302 that guide users through camera setup, positioning requirements, and movement calibration procedures necessary for accurate computer vision tracking. The progression can include setup screens 1304 that help users configure their environment (e.g., camera angle, lighting conditions, or space requirements) and verify that the computer vision system can properly detect their movements. The interface can display motion insight screens 1306 that provide near-real-time or real-time feedback about exercise form, movement quality, and performance metrics (e.g., repetition counting, range of motion analysis, or posture correction guidance) generated by the playlist generation platform, such as by the exercise controller 734.

[0126] Figure 14 illustrates a user interface 1400 that enables user interaction to receive user feedback. In some implementations, the user interface 1400 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different39183067633.1Attorney Docket No. 125847.8045. WO01 orders.

[0127] The user interface facilitates the collection of user responses and outcomes that inform the playlist generation platform's adaptive learning capabilities. The interface can present a session summary 1402 that displays information about completed activities, including exercise performance metrics, body region focus areas (e.g., shoulder, upper back, neck), and quantitative measures such as repetition counts or session duration. The interface can include a feedback interface 1404 that enables users to provide subjective assessments about activity difficulty, pain levels, or exercise effectiveness through interactive controls (e.g., rating scales, difficulty adjustments, or preference indicators). Additionally, the interface can display a feedback confirmation 1406 that acknowledges user input and explains how the provided feedback will be used to personalize future playlist recommendations and adjust therapeutic content intensity or focus areas.

[0128] Figure 15 illustrates a user interface 1500 that enables user interaction for a logging activity (here, a health logging activity) within the generated playlist. In some implementations, the user interface 1500 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0129] The user interface presents a health data collection activity that represents one of the polymorphic activity types within the playlist generation platform's domain structure. The interface can display pain scale labels 1502 that identify specific body regions (e.g., neck, back, shoulder) for which users can provide subjective health assessments. The interface can also include pain scale indicators 1504 that enable users to quantify their current pain levels or symptom severity using interactive slider controls or rating scales that range from minimal discomfort to maximum intensity. The health logging can be used as input for subsequent analysis of the user state by the playlist generation40183067633.1Attorney Docket No. 125847.8045. WO01 platform, and be used as retraining data for the routing machine learning model (e.g., the orchestration engine) and / or domain-specific models. For example, the routing machine learning model and / or domain-specific models can be retrained on activities labeled with an indicator of the health logging data (i.e., Activity A labeled with “too easy”).

[0130] Figure 16 illustrates a user interface 1600 that enables user interaction for an education activity within the generated playlist. In some implementations, the user interface 1600 is implemented by a system (e.g., the playlist generation platform) including components of the example processing system 1800 illustrated and described in more detail with reference to Figure 18. The system can be implemented on a terminal device, on a server, or on a telecommunications network core. Likewise, implementations can include different and / or additional operations or can perform the operations in different orders.

[0131] The user interface presents educational content that represents a particular polymorphic activity type of an activity presented by the playlist generation platform. The interface can display a timestamp indicator 1602 that shows the current time or session duration. The educational activity can include informational content about programs associated with the playlist, health conditions, treatment approaches (e.g., explanatory text, instructional graphics, or interactive learning modules), and so forth. This educational component may be selected by the education domain-specific model based on the user's current knowledge gaps, treatment stage, or objectives.Example Implementation of the Models of the Playlist Generation Platform

[0184] Figure 17 illustrates a layered architecture of an Al system 1700 that can implement the ML models of the playlist generation platform, in accordance with some implementations of the present technology. Example ML models can include the models executed by the playlist generation platform. Accordingly, the models of the playlist generation platform can include one or more components of the Al system 1700.

[0185] As shown, the Al system 1700 can include a set of layers, which conceptually organize elements within an example network topology for the Al41183067633.1Attorney Docket No. 125847.8045. WO01 system’s architecture to implement a particular Al model. Generally, an Al model is a computer-executable program implemented by the Al system 1700 that analyzes data to make predictions. Information can pass through each layer of the Al system 1700 to generate outputs for the Al model. The layers can include a data layer 1702, a structure layer 1704, a model layer 1706, and an application layer 1708. The algorithm 1716 of the structure layer 1704 and the model structure 1720 and model parameters 1722 of the model layer 1706 together form an example Al model. The loss function engine 1724, optimizer 1726, and regularization engine 1728 work to refine and optimize the Al model, and the data layer 1702 provides resources and support for application of the Al model by the application layer 1708.

[0186] The data layer 1702 acts as the foundation of the Al system 1700 by preparing data for the Al model. As shown, the data layer 1702 can include two sub-layers: a hardware platform 1710 and one or more software libraries 1712. The hardware platform 1710 can be designed to perform operations for the Al model and include computing resources for storage, memory, logic, and networking, such as the resources described in relation to Figures 17 and 15. The hardware platform 1710 can process amounts of data using one or more servers. The servers can perform backend operations such as matrix calculations, parallel calculations, ML training, and the like. Examples of servers used by the hardware platform 1710 include central processing units (CPUs) and graphics processing units (GPUs). CPUs are electronic circuitry designed to execute instructions for computer programs, such as arithmetic, logic, controlling, and input / output (I / O) operations, and can be implemented on integrated circuit (IC) microprocessors. GPUs are electric circuits that were originally designed for graphics manipulation and output but may be used for Al applications due to their vast computing and memory resources. GPUs use a parallel structure that generally makes their processing more efficient than that of CPUs. In some instances, the hardware platform 1710 can include computing resources (e.g., servers, memory, etc.) offered by a cloud services provider. The hardware platform 1710 can also include computer memory for storing data about the Al model, application of the Al model, and training data for the Al model. The computer memory can be a form of random-access42183067633.1Attorney Docket No. 125847.8045. WO01 memory (RAM), such as dynamic RAM, static RAM, and non-volatile RAM.

[0187] The software libraries 1712 can be thought of as suites of data and programming code, including executables, used to control the computing resources of the hardware platform 1710. The programming code can include low-level primitives (e.g., fundamental language elements) that form the foundation of one or more low-level programming languages, such that servers of the hardware platform 1710 can use the low-level primitives to carry out specific operations. The low-level programming languages do not require much, if any, abstraction from a computing resource’s instruction set architecture, enabling them to run quickly with a small memory footprint. Examples of software libraries 1712 that can be included in the Al system 1700 include INTEL Math Kernel Library, NVIDIA cuDNN, EIGEN, and OpenBLAS.

[0188] The structure layer 1704 can include an ML framework 1714 and an algorithm 1716. The ML framework 1714 can be thought of as an interface, library, or tool that enables users to build and deploy the Al model. The ML framework 1714 can include an open-source library, an API, a gradientboosting library, an ensemble method, and / or a deep learning toolkit that works with the layers of the Al system to facilitate development of the Al model. For example, the ML framework 1714 can distribute processes for application or training of the Al model across multiple resources in the hardware platform 1710. The ML framework 1714 can also include a set of pre-built components that have the functionality to implement and train the Al model and enable users to use pre-built functions and classes to construct and train the Al model. Thus, the ML framework 1714 can be used to facilitate data engineering, development, hyperparameter tuning, testing, and training for the Al model. Examples of ML frameworks 1714 that can be used in the Al system 1700 include TENSORFLOW, PYTORCH, SCIKIT-LEARN, KERAS, LightGBM, RANDOM FOREST, and AMAZON WEB SERVICES.

[0189] The algorithm 1716 can be an organized set of computer-executable operations used to generate output data from a set of input data and can be described using pseudocode. The algorithm 1716 can include complex code that enables the computing resources to learn from new input data and create43183067633.1Attorney Docket No. 125847.8045. WO01 new / modified outputs based on what was learned. In some implementations, the algorithm 1716 can build the Al model through being trained while running computing resources of the hardware platform 1710. This training enables the algorithm 1716 to make predictions or decisions without being explicitly programmed to do so. Once trained, the algorithm 1716 can run at the computing resources as part of the Al model to make predictions or decisions, improve computing resource performance, or perform tasks. The algorithm 1716 can be trained using supervised learning, unsupervised learning, semisupervised learning, and / or reinforcement learning.

[0190] Using supervised learning, the algorithm 1716 can be trained to learn patterns (e.g., map input data to output data) based on labeled training data. The training data may be labeled by an external user or operator. For instance, a user may collect a set of training data, such as by capturing data from sensors, images from a camera, outputs from a model, and the like. In an example implementation, training data can include native-format data collected (e.g., in the form of a video feed) from various source computing systems described in relation to Figures 1 -13. Furthermore, training data can include pre-processed data generated by various engines of the playlist generation platform described in relation to Figures 1 -13. The user may label the training data based on one or more classes and train the Al model by inputting the training data into the algorithm 1716. The algorithm 1716 determines how to label the new data based on the labeled training data. The user can facilitate collection, labeling, and / or input via the ML framework 1714. In some instances, the user may convert the training data to a set of feature vectors for input to the algorithm 1716. Once trained, the user can test the algorithm 1716 on new data to determine whether the algorithm 1716 is predicting accurate labels for the new data. For example, the user can use cross-validation methods to test the accuracy of the algorithm 1716 and retrain the algorithm 1716 on new training data if the results of the cross-validation are below an accuracy threshold.

[0191] Supervised learning can include classification and / or regression. Classification techniques include teaching the algorithm 1716 to identify a category of new observations based on training data and are used when input data for the algorithm 1716 is discrete. Said differently, when learning through44183067633.1Attorney Docket No. 125847.8045. WO01 classification techniques, the algorithm 1716 receives training data labeled with categories (e.g., classes) and determines how features observed in the training data (e.g., various claim elements, policy identifiers, tokens extracted from unstructured data) relate to the categories (e.g., risk propensity categories, claim leakage propensity categories, complaint propensity categories). Once trained, the algorithm 1716 can categorize new data by analyzing the new data for features that map to the categories. Examples of classification techniques include boosting, decision tree learning, genetic programming, learning vector quantization, k-nearest neighbor (k-NN) algorithm, and statistical classification.

[0192] Regression techniques include estimating relationships between independent and dependent variables and are used when input data to the algorithm 1716 is continuous. Regression techniques can be used to train the algorithm 1716 to predict or forecast relationships between variables. To train the algorithm 1716 using regression techniques, a user can select a regression method for estimating the parameters of the model. The user collects and labels training data that is input to the algorithm 1716 such that the algorithm 1716 is trained to understand the relationship between data features and the dependent variable(s). Once trained, the algorithm 1716 can predict missing historic data or future outcomes based on input data. Examples of regression methods include linear regression, multiple linear regression, logistic regression, regression tree analysis, least squares method, and gradient descent. In an example implementation, regression techniques can be used, for example, to estimate and fill in missing data for ML-based pre-processing operations.

[0193] Under unsupervised learning, the algorithm 1716 learns patterns from unlabeled training data. In particular, the algorithm 1716 is trained to learn hidden patterns and insights of input data, which can be used for data exploration or for generating new data. Here, the algorithm 1716 does not have a predefined output, unlike the label’s output when the algorithm 1716 is trained using supervised learning. Said another way, unsupervised learning is used to train the algorithm 1716 to find an underlying structure of a set of data, group the data according to similarities, and represent that set of data in a compressed format. The playlist generation platform can use unsupervised45183067633.1Attorney Docket No. 125847.8045. WO01 learning to identify patterns in claim history (e.g., to identify particular event sequences) and so forth. In some implementations, performance of the models of the playlist generation platform that can use unsupervised learning is improved because the incoming video feed is pre-processed and reduced, based on the relevant triggers, as described herein.

[0194] A few techniques can be used in supervised learning: clustering, anomaly detection, and techniques for learning latent variable models. Clustering techniques include grouping data into different clusters that include similar data, such that other clusters contain dissimilar data. For example, during clustering, data with possible similarities remains in a group that has fewer or no similarities to another group. Examples of clustering techniques include density-based methods, hierarchical-based methods, partitioning methods, and grid-based methods. In one example, the algorithm 1716 may be trained to be a k-means clustering algorithm, which partitions n observations in k clusters such that each observation belongs to the cluster with the nearest mean serving as a prototype of the cluster. Anomaly detection techniques are used to detect previously unseen rare objects or events represented in data without prior knowledge of these objects or events. Anomalies can include data that occur rarely in a set, a deviation from other observations, outliers that are inconsistent with the rest of the data, patterns that do not conform to well- defined normal behavior, and the like. When using anomaly detection techniques, the algorithm 1716 may be trained to be an Isolation Forest, local outlier factor (LOF) algorithm, or k-nearest neighbor (k-NN) algorithm. Latent variable techniques include relating observable variables to a set of latent variables. These techniques assume that the observable variables are the result of an individual’s position on the latent variables and that the observable variables have nothing in common after controlling for the latent variables. Examples of latent variable techniques that may be used by the algorithm 1716 include factor analysis, item response theory, latent profile analysis, and latent class analysis.

[0195] The model layer 1706 implements the Al model using data from the data layer 1702 and the algorithm 1716 and ML framework 1714 from the structure layer 1704, thus enabling decision-making capabilities of the Al46183067633.1Attorney Docket No. 125847.8045. WO01 system 1700. The model layer 1706 includes a model structure 1720, model parameters 1722, a loss function engine 1724, an optimizer 1726, and a regularization engine 1728.

[0196] The model structure 1720 describes the architecture of the Al model of the Al system 1700. The model structure 1720 defines the complexity of the pattern / relationship that the Al model expresses. Examples of structures that can be used as the model structure 1720 include decision trees, support vector machines, regression analyses, Bayesian networks, Gaussian processes, genetic algorithms, and artificial neural networks (or, simply, neural networks). The model structure 1720 can include a number of structure layers, a number of nodes (or neurons) at each structure layer, and activation functions of each node. Each node’s activation function defines how the node converts data received to data output. The structure layers may include an input layer of nodes that receive input data and an output layer of nodes that produce output data. The model structure 1720 may include one or more hidden layers of nodes between the input and output layers. The model structure 1720 can be an artificial neural network (or, simply, neural network) that connects the nodes in the structured layers such that the nodes are interconnected. Examples of neural networks include feedforward neural networks, CNNs, recurrent neural networks (RNNs), autoencoder, and generative adversarial networks (GANs).

[0197] The model parameters 1722 represent the relationships learned during training and can be used to make predictions and decisions based on input data. The model parameters 1722 can weigh and bias the nodes and connections of the model structure 1720. For instance, when the model structure 1720 is a neural network, the model parameters 1722 can weight and bias the nodes in each layer of the neural networks, such that the weights determine the strength of the nodes and the biases determine the thresholds for the activation functions of each node. The model parameters 1722, in conjunction with the activation functions of the nodes, determine how input data is transformed into desired outputs. The model parameters 1722 can be determined and / or altered during training of the algorithm 1716.

[0198] The loss function engine 1724 can determine a loss function, which47183067633.1Attorney Docket No. 125847.8045. WO01 is a metric used to evaluate the Al model’s performance during training. For instance, the loss function engine 1724 can measure the difference between a predicted output of the Al model and the actual output of the Al model and is used to guide optimization of the Al model during training to minimize the loss function. The loss function may be presented via the ML framework 1714, such that a user can determine whether to retrain or otherwise alter the algorithm 1716 if the loss function is over a threshold. In some instances, the algorithm 1716 can be retrained automatically if the loss function is over the threshold. Examples of loss functions include a binary-cross entropy function, hinge loss function, regression loss function (e.g., mean square error, quadratic loss, etc.), mean absolute error function, smooth mean absolute error function, log-cosh loss function, and quantile loss function.

[0199] The optimizer 1726 adjusts the model parameters 1722 to minimize the loss function during training of the algorithm 1716. In other words, the optimizer 1726 uses the loss function generated by the loss function engine 1724 as a guide to determine what model parameters lead to the most accurate Al model. Examples of optimizers include Gradient Descent (GD), Adaptive Gradient Algorithm (AdaGrad), Adaptive Moment Estimation (Adam), Root Mean Square Propagation (RMSprop), Radial Base Function (RBF), and Limited-memory BFGS (L-BFGS). The type of optimizer 1726 used may be determined based on the type of model structure 1720 and the size of data and the computing resources available in the data layer 1702.

[0200] The regularization engine 1728 executes regularization operations. Regularization is a technique that prevents over- and underfitting of the Al model. Overfitting occurs when the algorithm 1716 is overly complex and too adapted to the training data, which can result in poor performance of the Al model. Underfitting occurs when the algorithm 1716 is unable to recognize even basic patterns from the training data such that it cannot perform well on training data or on validation data. The optimizer 1726 can apply one or more regularization techniques to fit the algorithm 1716 to the training data properly, which helps constrain the resulting Al model and improves its ability for generalized applications. Examples of regularization techniques include lasso (L1 ) regularization, ridge (L2) regularization, and elastic (L1 and L2)48183067633.1Attorney Docket No. 125847.8045. WO01 regularization.

[0201] The application layer 1708 describes how the Al system 1700 is used to solve problems or perform tasks. In an example implementation, the application layer 1708 can include a front-end user interface of the playlist generation platform.Example Computing Environment of the Playlist Generation Platform

[0202] Figure 18 includes a block diagram illustrating an example of a processing system 1800 in which at least some operations described herein can be implemented. For example, components of the processing system 1800 may be hosted on a computing device that includes a playlist generation platform (e.g., playlist generation platform 202 of Figure 2 or playlist generation platform 312 of Figure 3).

[0203] The processing system 1800 can include a processor 1802, main memory 1806, non-volatile memory 1810, network adapter 1812, video display 1818, input / output devices 1820, control device 1822 (e.g., a keyboard or pointing device such as a computer mouse or trackpad), drive unit 1824 including a storage medium 1826, and signal generation device 1830 that are communicatively connected to a bus 1816. The bus 1816 is illustrated as an abstraction that represents one or more physical buses or point-to-point connections that are connected by appropriate bridges, adapters, or controllers. The bus 1816, therefore, can include a system bus, a Peripheral Component Interconnect (“PCI”) bus or PCI-Express bus, a HyperTransport (“HT”) bus, an Industry Standard Architecture (“ISA”) bus, a Small Computer System Interface (“SCSI”) bus, a Universal Serial Bus (“USB”) data interface, an Inter-Integrated Circuit (“I2C”) bus, or a high-performance serial bus developed in accordance with Institute of Electrical and Electronics Engineers (“IEEE”) 1394.

[0204] While the main memory 1806, non-volatile memory 1810, and storage medium 1826 are shown to be a single medium, the terms “machine- readable medium” and “storage medium” should be taken to include a single medium or multiple media (e.g., a centralized / distributed database and / or associated caches and servers) that store one or more sets of instructions 1828. The terms “machine-readable medium” and “storage medium” shall also49183067633.1Attorney Docket No. 125847.8045. WO01 be taken to include any medium that is capable of storing, encoding, or carrying a set of instructions for execution by the processing system 1800.

[0205] In general, the routines executed to implement the embodiments of the disclosure can be implemented as part of an operating system or a specific application, component, program, object, module, or sequence of instructions (collectively referred to as “computer programs”). The computer programs typically comprise one or more instructions (e.g., instructions 1804, 1808, 1828) set at various times in various memory and storage devices in a computing device. When read and executed by the processor 1802, the instruction(s) cause the processing system 1800 to perform operations to execute elements involving the various aspects of the present disclosure.

[0206] Further examples of machine- and computer-readable media include recordable-type media, such as volatile memory devices and non-volatile memory 1810 devices, removable disks, hard disk drives, and optical disks (e.g., Compact Disk Read-Only Memory (“CD-ROMs”) and Digital Versatile Disks (“DVDs”)), and transmission-type media, such as digital and analog communication links.

[0207] The network adapter 1812 enables the processing system 1800 to mediate data in a network 1814 with an entity that is external to the processing system 1800 through any communication protocol supported by the processing system 1800 and the external entity. The network adapter 1812 can include a network adaptor card, a wireless network interface card, a router, an access point, a wireless router, a switch, a multilayer switch, a protocol converter, a gateway, a bridge, bridge router, a hub, a digital media receiver, a repeater, or any combination thereof.Remarks

[0208] The foregoing description of various embodiments of the claimed subject matter has been provided for the purposes of illustration and description. It is not intended to be exhaustive or to limit the claimed subject matter to the precise forms disclosed. Many modifications and variations will be apparent to one skilled in the art. Embodiments were chosen and described in order to best describe the principles of the invention and its practical50183067633.1Attorney Docket No. 125847.8045. WO01 applications, thereby enabling those skilled in the relevant art to understand the claimed subject matter, the various embodiments, and the various modifications that are suited to the particular uses contemplated.

[0209] Although the Detailed Description describes certain embodiments and the best mode contemplated, the technology can be practiced in many ways no matter how detailed the Detailed Description appears. Embodiments can vary considerably in their implementation details, while still being encompassed by the specification. Particular terminology used when describing certain features or aspects of various embodiments should not be taken to imply that the terminology is being redefined herein to be restricted to any specific characteristics, features, or aspects of the technology with which that terminology is associated. In general, the terms used in the following claims should not be construed to limit the technology to the specific embodiments disclosed in the specification, unless those terms are explicitly defined herein. Accordingly, the actual scope of the technology encompasses not only the disclosed embodiments, but also all equivalent ways of practicing or implementing the embodiments.

[0210] The language used in the specification has been principally selected for readability and instructional purposes. It may not have been selected to delineate or circumscribe the subject matter. It is therefore intended that the scope of the technology be limited not by this Detailed Description, but rather by any claims that issue on an application based hereon. Accordingly, the disclosure of various embodiments is intended to be illustrative, but not limiting, of the scope of the technology as set forth in the following claims.51183067633.1

Claims

Attorney Docket No. 125847.8045. WO01CLAIMSWhat is claimed is:1 . A method for generating personalized activity sequences performed by a computer program executing on a computing device, the method comprising: receiving, via a network interface coupled to the computing device, user state data of a user from a plurality of data sources; accessing an activities repository that identifies a plurality of activities for the user across multiple activity domains, wherein each activity corresponds to one or more activity domains, and wherein one or more activities of the plurality of activities are represented by different data types; applying, to the user state data and the multiple activity domains, a routing machine learning model trained to map each portion of the user state data to a subset of activity domains within the multiple activity domains based on predefined information of the user and other portions of the user state data; for each domain within the subset of activity domains, applying, to the user state data and respective activities of the activity domain, a respective domain-specific machine learning model associated with the activity domain that is trained to generate one or more candidate activities based on the user state data; generating a polymorphic type data structure by aggregating and indexing the one or more candidate activities for each domain across the subset of activity domains; and executing, via the network interface, the candidate activities indexed within the polymorphic type data structure by, for each candidate activity, transmitting computer-executable instructions to the computing device based on a respective data type of the candidate activity.

2. The method of claim 1 , wherein the user state data includes at least one of: physiological sensor data,52183067633.1Attorney Docket No. 125847.8045. WO01 user-reported survey responses, historical activity completion data, or real-time location information.

3. The method of claim 1 , wherein the activities repository further comprises metadata for each activity, the metadata including at least one of: a difficulty level, an estimated duration, an equipment list, or a target body region.

4. The method of claim 1 , wherein the polymorphic type data structure is an array or list in which each element includes a reference to a candidate activity and an identifier of the respective data type of the candidate activity.

5. The method of claim 1 , wherein executing the candidate activities further comprises: for each candidate activity, identifying a user interface modality to present the activity based on the respective data type of the candidate activity.

6. The method of claim 1 , further comprising: prior to executing the candidate activities, filtering the candidate activities based on a predefined set of user constraints.

7. The method of claim 1 , further comprising: receiving, subsequent to execution of at least one candidate activity, user feedback data indicating a response to the activity; and storing the user feedback data in association with the corresponding activity and user state data.53183067633.1Attorney Docket No. 125847.8045. WO018. The method of claim 1 , wherein each domain-specific machine learning model is trained on different training datasets.

9. The method of claim 1 , wherein the routing machine learning model is trained on a training dataset that includes labeled associations between reference user state data and activity domains.

10. A non-transitory medium with instructions stored thereon that, when executed by a processor of a computing device, cause the computing device to perform operations comprising: obtaining, via a network interface coupled to the computing device, user state information of a user; identify, from a repository, a plurality of items, each item being associated with at least one of a plurality of item domains and represented by one of a plurality of data types; determine, using a routing artificial intelligence model, a subset of item domains for the user by correlating portions of the user state information with one or more of predefined user attributes or other portions of the user state information; for each item domain in the subset of item domains, trigger a corresponding domain-specific artificial intelligence model to identify one or more candidate items from the items within the repository associated with a respective domain of the corresponding domain-specific artificial intelligence model; assemble a polymorphic data structure by indexing the candidate items from each item domain in the subset of item domains; and for each candidate item in the polymorphic data structure, cause transmission of executable instructions, via the network interface, to the computing device, based on a respective data type of the candidate item.

11. The non-transitory medium of claim 10, wherein the operations further comprise:54183067633.1Attorney Docket No. 125847.8045. WO01 prioritizing the candidate items indexed within the polymorphic data structure based on a ranking score for each candidate item, wherein the ranking score is determined based on one or more of the user state information or historical user interaction data.

12. The non-transitory medium of claim 10, wherein the operations further comprise: generating a report that summarizes the data types and item domains of the candidate items over a particular time period.

13. The non-transitory medium of claim 10, wherein the operations further comprise: updating at least one of the routing artificial intelligence model or a domain-specific artificial intelligence model based on received user feedback data.

14. The non-transitory medium of claim 10, wherein the operations further comprise: filtering the plurality of items in the repository based on at least one user- specified exclusion criterion prior to determining the subset of item domains.

15. A non-transitory medium with instructions stored thereon that, when executed by a processor of a computing device, cause the computing device to perform operations comprising: obtaining (a) user state information of a user and (b) a repository identifying a plurality of items, wherein each item is associated with at least one of a plurality of item domains and represented by one of a plurality of data types; generating a first polymorphic data structure by triggering a series of domain-specific artificial intelligence models to identify a first set of items within the repository based on a respective item domain of each item within the repository;55183067633.1Attorney Docket No. 125847.8045. WO01 for each item in the first polymorphic data structure, cause transmission of a first set of executable instructions to the computing device, based on a respective data type of the identified first set of items; receiving, via the computing device, interaction data responsive to the transmission of the first set of executable instructions; adjusting one or more model parameters of one or more domain-specific artificial intelligence models based on the interaction data; generating a second polymorphic data structure by triggering the series of domain-specific artificial intelligence models to identify a second set of items within the repository based on a respective item domain of each item within the repository and the adjusted one or more parameters; and for each item in the second polymorphic data structure, cause transmission of a second set of executable instructions to the computing device, based on a respective data type of the identified second set of items.

16. The non-transitory medium of claim 15, wherein the operations further comprise: storing the interaction data in a user profile database, wherein the interaction data is indexed by a particular item domain and a particular data type.

17. The non-transitory medium of claim 15, wherein the first set of executable instructions and the second set of executable instructions are each configured to trigger a different user interface presentation based on a respective data type of the respective item.

18. The non-transitory medium of claim 15, wherein the operations further comprise: generating a notification configured to be presented on the computing device that indicates that the second set of items is generated based on the interaction data.56183067633.1Attorney Docket No. 125847.8045. WO0119. The non-transitory medium of claim 15, wherein the operations further comprise: logging the adjustment to the one or more model parameters in a database.

20. The non-transitory medium of claim 15, wherein the operations further comprise: training at least one domain-specific artificial intelligence model using respective interaction data associated with multiple users.57183067633.1

Images