Adaptive response generation and asynchronous multi¬ device media objects management

Abstract

Apparatus and related methods relate to adaptive generation of transformed multimedia content using predefined prompts trained from user preferences and past interactions. In an illustrative embodiment, a computer-implemented method may include receiving a signal to generate an adaptively transformed content. The method may, for example, classify multimedia content into a record type. The method may, for example, select a predefined prompt based on the record type and a user profile. The method may, for example, apply a large language model with the predefined prompt to generate the adaptively transformed content. The method may, for example, receive user feedback and apply it to a user preference learning model to produce prompt-adjustment parameters. The method may, for example, update the user profile or predefined prompt for later use without retraining the large language model. Various embodiments may advantageously enable cost-effective and dynamic predefined prompt adaptation for transforming multimedia content.

Description

Docket #: 1040-05WO / USADAPTIVE RESPONSE GENERATION AND ASYNCHRONOUS MULTIDEVICE MEDIA OBJECTS MANAGEMENT CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Application Serial No.63 / 721,248, titled “Adaptive Response Generation and Asynchronous Multi -Device Media Objects Management,” filed by Gregory Odland, et al., on November 15, 2024. This application incorporates the entire contents of the foregoing application(s) herein by reference.

[0001] This application may share inventor(s) and / or subject matter with one or more of the following applications: US 16 / 876,083, filed May 17, 2020 titled "Apparatus for generating and transmitting annotated video sequences in response to manual and image input devices"; PCT / US2020 / 033328, filed May 17, 2020 titled "Apparatus for generating and transmitting annotated video sequences in response to manual and image input devices"; PCT / US2022 / 072888, filed Jun 10, 2022 titled "Multi-party controlled transient user credentialing for interaction with patient health data"; EP 22738836.0, filed Jun 10, 2022 titled "Multi-party controlled transient user credentialing for interaction with patient health data"; US 17 / 806,446, filed Jun 10, 2022 (naming inventor(s) including ODLAND, Gregory) and titled "Multi-party controlled transient user credentialing for interaction with patient health data"; US 18 / 340,319, filed Jun 23, 2023 titled "Multi-party controlled transient user credentialing for interaction with patient health data"; US 19 / 081,300, filed Mar 17, 2025 titled "Multi-party controlled transient user credentialing for interaction with patient health data"; PCT / US2024 / 035272, filed Jun 24, 2024 titled "Dynamically generated LLM request package"; US 18 / 752,396, filed Jun 24, 2024 titled "Dynamically generated LLM request package"; and US 63 / 510,012, filed Jun 23, 2023 titled "Dynamically generated LLM request package." The entire contents of each of the foregoing applications and their priority applications, if any, are incorporated herein by reference.

[0002] Unless expressly stated, changes in terminology from priority application(s) to this application are made without prejudice or disclaimer of subject matter. Changes from the priority application(s) (e.g., provisional applications(s)) are intended to be broadening and / or additive unless expressly stated otherwise. Replacement of alternative terms with a single representative term, for example, are inclusive unless otherwise defined. Various embodiments may also be found in previous disclosure(s) incorporated by reference.Embodiments of similar languages in this application are not modifications or disclaimer ofDocket #: 1040-05WO / USthe embodiments disclosed in previous incorporated disclosures unless otherwise stated.TECHNICAL FIELD

[0002] Various embodiments relate generally to automatic note compilation and / or generation systems.BACKGROUND

[0003] Digital Medical Records (DMRs), commonly known as Electronic Medical Records (EMRs) or Electronic Health Records (EHRs), have transformed healthcare by digitizing patient information, making it more accessible and manageable. DMRs encompass a patient's medical history, including diagnoses, medications, immunization dates, allergies, lab results, and treatment plans, all stored in a digital format. By enabling healthcare providers to access up-to-date patient information quickly, DMRs may improve care coordination, enhance treatment accuracy, and / or reduce medical errors.

[0004] In one example, generating digital medical records may include data entry during patient intake. For example, patient information may be recorded by a healthcare provider into an EMR system. Data may, in some cases, be directly collected from diagnostic devices and / or medical equipment. For example, the diagnostic devices may generate measurement data related to a patient’s health conditions. For example, digital health tools like wearable devices, telemedicine platforms, and patient portals may contribute real-time data on vital signs, activity levels, and / or other health metrics.

[0005] Digital records may be securely stored in centralized databases and / or cloud-based systems, in some examples. For example, the databases may be protected by encryption and / or security protocols (e.g., Health Insurance Portability and Accountability Act (HIPAA) compliant security protocols). Authorized healthcare providers may retrieve these records for follow-up visits, specialist consultations, or emergency care, accessing crucial patient data instantly. Efficient retrieval may, for example, aid in patient continuity of care. For example, providers may review past diagnoses and treatments for facilitating timely and informed medical decision-making.SUMMARY

[0006] Apparatus and related methods relate to adaptive generation of transformed multimedia content using predefined prompts trained from user preferences and past interactions. In an illustrative embodiment, a computer-implemented method may includeDocket #: 1040-05WO / USreceiving a signal to initiate generation of an adaptively transformed content. The method may, for example, include classifying multimedia content into a record type. The method may, for example, include selecting a predefined prompt based on the record type and a user profile. The method may, for example, include applying a large language model with the predefined prompt to generate the adaptively transformed content. The method may, for example, include receiving user feedback and applying the feedback to a user preference learning model to produce prompt-adjustment parameters. The method may, for example, include updating the user profile or predefined prompt for later use without retraining the large language model. Various embodiments may advantageously enable cost-effective and dynamic prompt adaptation for multimedia transformation.

[0007] Apparatus and related methods relate to asynchronous integration of multimedia content into output content by identifying associations between input and stored content. In an illustrative embodiment, a system may include a processor configured to asynchronously update electronic content. The processor may, for example, receive asynchronous input content related to an existing record type. The processor may, for example, identify a stored record associated with the asynchronous input content. The processor may, for example, generate a transformed content based on both the asynchronous input content and the stored record. The processor may, for example, integrate the transformed content with the stored record to create an updated integrated content. The processor may, for example, determine a target destination for the updated integrated content based on an association found in the input content. The processor may, for example, transmit the updated integrated content to the target destination. Various embodiments may advantageously enable automatic, incremental, and asynchronous construction of medical records.

[0008] Various embodiments may achieve one or more advantages. For example, some embodiments may advantageously enable user-level adaptability without retraining a large language model. Some embodiments, for example, may advantageously improve energy efficiency and computational performance. For example, some embodiments may advantageously provide cost-effective and dynamic adaptation of predefined prompts. Some embodiments, for example, may advantageously enhance personalization and responsiveness of automated content generation. For example, some embodiments may advantageously improve continuity and traceability of asynchronous workflows. Some embodiments, for example, may advantageously reduce manual routing errors and improve workflow efficiency. For example, some embodiments may advantageously facilitate incremental andDocket #: 1040-05WO / USasynchronous building of electronic records. Some embodiments, for example, may advantageously improve user experience and data frequency across multiple contributors and devices.

[0009] The details of various embodiments are set forth in the accompanying drawings and the description below. Other features and advantages will be apparent from the description and drawings, and from the claims.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] FIG. 1 depicts an exemplary adaptive content generation system (ACGS) employed in an illustrative use-case scenario.

[0011] FIG. 2 depicts an exemplary continuous content generation system (CCGS) employed in an illustrative use-case scenario.

[0012] FIG. 3 is a block diagram depicting an exemplary ACGS.

[0013] FIG. 4 is a flowchart illustrating an exemplary interactive electronic medical record generation method.

[0014] FIG. 5 is a flowchart illustrating an exemplary continuous medical record integration method.

[0015] FIG. 6 shows an example predefined prompt training system.

[0016] FIG. 7 shows an example association data store.

[0017] Like reference symbols in the various drawings indicate like elements.DETAILED DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS

[0018] FIG. 1 depicts an exemplary adaptive content generation system (ACGS 100) employed in an illustrative use-case scenario. As shown, the ACGS 100 receives a multimedia content 105. For example, the multimedia content 105 may be a clinical note received from a user 110 (e.g., a medical personnel). For example, the multimedia content 105 may include an image. For example, the multimedia content 105 may include a voice note. For example, the multimedia content 105 may include a video.

[0019] The ACGS 100 includes a note generation engine (NGE 115) to receive the multimedia content 105. For example, the NGE 115 may include a note classifier 120 to classify the multimedia content 105. For example, the NGE 115 may apply a note editing large language model (note editing LLM 125) to the multimedia content 105 to generate a transformed content 135 (e.g., a completed clinical note) to a user device 130 of the user 110. In some implementations, the NGE 115 may insert protected health information (PHI) (e.g., patientDocket #: 1040-05WO / USname, patient’s medical history, patient’s identification number) into the transformed content 135 based on an authentication rights of the user 110.

[0020] For example, the NGE 115 may apply one or more predefined prompts to edit the multimedia content 105. For example, the predefined prompt 140 may request the NGE 115 to add more detail. For example, the predefined prompt 140 may instruct the NGE 115 to change a name of the patient. For example, the predefined prompt 140 may command the NGE 115 to convert the clinical note into another language. In some implementations, the NGE 115 may automatically (dynamically) generate the predefined prompt 140 based on a user profile 145 of the user and a classification of the multimedia content 105 (e.g., by applying one or more Al models).

[0021] The predefined prompt 140, for example, may include artificial intelligence (Al) generated suggestions in editing the multimedia content 105. As shown, the user device 130 may transmit a predefined prompt 140 (e.g., as input from the user 110) to the NGE 115 edit the multimedia content 105.

[0022] As an illustrative example, the user 110 may record a voice note to be transmitted to the ACGS 100. For example, upon receiving the voice note, the NGE 115 may classify it as, for example, an after-visit summary. For example, the NGE 115 may transcribe the voice note and generate a clinical note. For example, the NGE 115 may apply the LLM 125 to modify the clinical note using the predefined prompt 140. For example, the predefined prompt 140 may include an “add more detail” instruction to instruct the LLM 125 to expand the clinical note with additional information.

[0023] In some implementations, the ACGS 100 may adaptively update the predefined prompt 140 based on user preferences and / or historical user behavior. As shown, the NGE 115 includes a user preference learning module (UPLM 150). For example, if a user frequently adds detailed information to a clinical note, the UPLM 150 may update the user profile 145 (e.g., of user specific weightings) to automatically include the predefined prompt 140 to add more detail when a clinical note is received in the future.

[0024] As shown, the UPLM 150 is operably coupled to the user device 130. For example, the UPLM 150 may directly receive input from the user 110 to update the transformed content 135. For example, the user 110 may generate a user-initiated feedback 155 to ask for details to be added in selected parts of the transformed content 135. The UPLM 150 may update the predefined prompt 140 (e.g., to identify relevant content and / or specific field to be expanded) associated with the user profile 145 based on the user-initiated feedback 155.Docket #: 1040-05WO / US

[0025] In some implementations, the UPLM 150 may engage with the user 110 via the user device 130 to update the predefined prompt 140. For example, the UPLM 150 may transmit a confirmation message 160 to the user device 130 (e.g., “Do you want to add more details in your future notes?”). For example, the user device 130 may answer the confirmation message 160 with a confirmation or with additional explanation and / or instructions. For example, the confirmation message 160 may include rules (e.g., “I don’t want any name in this part of notes because the name is already in another field.”). In some examples, the UPLM 150 may suggest edits (e.g., details of missing data) to the user device 130 via the confirmation message 160.

[0026] In various implementations, the content generation system (e.g., the ACGS 100) is configured to transform input content (e.g., the multimedia content 105) into target content (e.g., the transformed content 135) using automatically generated predefined prompts (e.g., the predefined prompt 140). The predefined prompt may be dynamically updated through a lightweight user-preference learning layer (e.g., the UPLM 150) based on identified user preferences (e.g., derived from the user-initiated feedback 155 received from a user) and a user profile (e.g., the user profile 145). The content generation system may advantageously provide a periodically updating (e.g., continuously improving) personalized artificial intelligence that adapts over time without retraining the underlying large language model, thereby reducing computational cost while maintaining individualized performance.

[0027] In conventional language model personalization, for example, each adjustment to model behavior may typically rely on retraining or fine-tuning a LLM (e.g., the editing LLM 125) with extensive datasets and high-compute resources. Some retraining processes may, for example, computationally expensive (e.g., in computational resources, energy) and / or timeconsuming. For example, continual improvement and redeployment are practically impossible. For example, in some systems, user preferences may be aggregated and re-applied through full or partial retraining cycles that may take months to complete in order to achieve personalization. In some embodiments, the ACGS 100 may advantageously improve the functioning of content transformation systems by dynamically adjusting the predefined prompt 140 adaptively based on user-initiated feedback (e.g., the user-initiated feedback 155) without retraining the LLM 125. Various embodiments may advantageously improve energy efficiency and user experience. For example, the ACGS 100 may provide a practical improvement in computer functionality and machine learning technology by achieving userlevel adaptability without the computational burden of traditional model retraining.Docket #: 1040-05WO / US

[0028] FIG. 2 depicts an exemplary continuous content generation system (CCGS 200) employed in an illustrative use-case scenario. For example, the CCGS 200 may allow multiple users 205 to collaborate on a single clinical note across different devices. In this example, a medical assistant 205a can start a note. For example, a doctor 205b may continue to edit the same note later with the doctor’s observations.

[0029] As shown, the CCGS 200 receives a data input 210 from the multiple users 205. For example, the multiple users 205 may include a human user. For example, the multiple users 205 may include a machine user. The CCGS 200 may generate an integrated content 215 based on the data input 210. The data input 210 may be generated by the medical assistant 205a and / or the doctor 205b. In some embodiments, the data input 210 may be generated from recording devices 205c (e.g., a voice recorder, a camera, a medical scanner including x-ray machine, a fluoroscopy machine, an MRI scanner, other sensors including weight measurement devices, blood pressure measurement devices). The data input 210 may also be generated from a chat 205d. For example, the chat 205d may include an Al chat with medical personnel and / or a patient.

[0030] As an illustrative example, the multiple users 205 may each record their part of a clinical note separately. The CCGS 200 includes a data integration engine 220 to merge the data input 210 from each of the multiple users 205 into an integrated content (e.g., a single cohesive note). For example, each of the multiple users 205 may use the user device 130 to generate notes and / or instructions to the CCGS 200. In some implementations, the data integration engine 220 may support asynchronous collaboration 225. For example, each of the multiple users 205 may asynchronously retrieve the integrated content 215 and contribute to the integrated content 215 at different times.

[0031] The CCGS 200 includes an integrated notes delivery engine (INDE 230) and an association data store 235. For example, the INDE 230 may deliver the integrated content 215 to one of the multiple users 205 based on a determined association and current situation based on the association data store 235. In some implementations, the integrated content 215 may include the transformed content 135. For example, a note may be started by the medical assistant 205a when a patient registers at the front desk. For example, the INDE 230 may, upon determining the type of a visit and an identity of the patient, transmit the note to the doctor 205b. As shown, the CCGS 200 is coupled to a patient history database 240. For example, the INDE 230 may, upon registration of the patient, populate a note based on a medical history of the patient stored in the patient history database 240.Docket #: 1040-05WO / US

[0032] In some examples, the CCGS 200 may begin with data received from the chat 205d with a registering patient (e.g., “why are you here? How are you feeling?”). For example, the medical assistant 205a may asynchronously receive the integrated content 215 containing preliminary data from the patient. For example, the medical assistant 205a may determine, based on the received integrated content 215, add more observations and / or details, about the patient. For example, the data integration engine 220 may also integrate audio clips and images of the patient before delivering the note to the doctor 205b. The doctor, for example, may review the received integrated content 215 and add additional information upon seeing the patient. Various embodiments may advantageously allow incremental and asynchronous building of a medical record automatically.

[0033] In some implementations, the association data store 235 may store multiple layers of relationships and / or contextual information. For example, the association data store 235 may include work relationships (e.g., between a nurse, a doctor, and a specialist involved in a same case). For example, the association data store 235 may include workflow information (e.g., intake, diagnosis, follow-up, consultation schedule). For example, the association data store 235 may include association rules defining how the INDE 230 may apply the relationships.

[0034] For example, the association data store 235 may include record-type correlations. For example, the association data store 235 may include user role definitions. For example, the association data store 235 may include organizational chart (e.g., departmental hierarchies).

[0035] For example, the association data store 235 may include (e.g., rule-based) event triggers that identify when and how the data integration engine 220 may transform the data input 210. For example, a rule may specify that once a registration note is finalized, the transformed content is automatically delivered to a physician scheduled for consultation associated with a patient related to the registration note. In some examples, the association data store 235 may include workflow rules linking imaging data from a technician to a radiologist review stage. As an illustrative example without limitation, the workflow rules may connect lab results to a pharmacist verification queue. In some examples, the INDE 230 may advantageously use the association data store 235 to facilitate context-aware transformation and distribution of the integrated content.

[0036] In some implementations, the INDE 230 may apply one or more of the predefined prompt 140 based on association information stored in the association data store 235 to determine an appropriate association, current situation, and / or target destination of the integrated content 215. For example, the INDE 230 may use the predefined prompt 140 toDocket #: 1040-05WO / USinterpret contextual information within the data input 210. For example, the INDE 230 may use the predefined prompt 140 to identify a type of the data input 210. For example, the INDE 230 may use the predefined prompt 140 to identify an associated appointment in the association data store 235 and the data input 210.

[0037] For example, the predefined prompt 140 may be dynamically refined based on user-initiated feedback 155 received. For example, if a doctor frequently requests additional summary data before signing off on a patient’s chart, the CCGS 200 may update the UPLM 150 to automatically expand relevant sections in future notes. As an illustrative example without limitation, if a pharmacist regularly corrects dosage instructions in integrated records, the predefined prompt 140 may adapt to highlight dosage fields before transmitting the transformed content for ease of the pharmacist. Various embodiments may advantageously provide efficient personalized improvements without retraining a large language model.

[0038] In various implementations, a continuous content generation system (e.g., the CCGS 200) may be configured to asynchronously integrate input content (e.g., the data input 210 from multiple users 205 into a unified integrated content 215. The CCGS 200 may, for example, enable collaboration across different workflow stages by associating each input with corresponding records based on relationships stored in the association data store 235. The system may, for example, apply the predefined prompt 140 to determine workflow context and transform received content before distribution to an appropriate target destination through the INDE 230. Various embodiments may advantageously extend the adaptive transformation process described with reference to FIG. 1 across multiple users and devices, thereby enabling context-aware and personalized updates without retraining the LLM 125 and improving system efficiency across asynchronous interactions.

[0039] In various implementations, a continuous content generation system (e.g., the CCGS 200) may be configured to asynchronously integrate data input 210 from multiple users 205 into a unified integrated content 215. The CCGS 200 may, for example, enable collaboration across different workflow stages by associating each input with corresponding records based on relationships stored in the association data store 235. The CCGS may, for example, apply predefined prompts (e.g., the predefined prompt 140) to determine workflow context and transform received content before distribution to an appropriate target destination through the INDE 230.

[0040] FIG. 3 is a block diagram depicting an exemplary ACGS 300. For example, the exemplary ACGS 300 may receive asynchronous notes from multiple user devices. ForDocket #: 1040-05WO / USexample, the exemplary ACGS 300 may apply the predefined prompt 140 to edit the received notes based on user’s preferences. In this example, the exemplary ACGS 300 includes a processor 305. The processor 305 may, for example, include one or more processing units. The processor 305 is operably coupled to a communication module 310. The communication module 310 may, for example, include wired communication. The communication module 310 may, for example, include wireless communication. In the depicted example, the communication module 310 is operably coupled to the user device 130, the multiple users 205, and the patient history database 240. For example, the exemplary ACGS 300 may receive notes and / or other multimedia content from the multiple users 205 asynchronously. In some implementations, the exemplary ACGS 300 may generate the transformed content 135 to the multiple users 205 and / or the user device 130. For example, the received instructions from the user device 130 for modifying the transformed content 135. For example, the exemplary ACGS 300 may assess the patient history database 240 to modify the multimedia content 105 received from the user device 130.

[0041] The processor 305 is operably coupled to a memory module 320. The memory module 320 may, for example, include one or more memory modules (e.g., random-access memory (RAM)). The processor 305 includes a storage module 325. The storage module 325 may, for example, include one or more storage modules (e.g., non-volatile memory). In the depicted example, the storage module 325 includes the NGE 115, the data integration engine 220, and the INDE 230. In this example, the storage module 325 includes a user preference learning engine (UPLE 330). For example, the UPLE 330 may include the UPLM 150.

[0042] The processor 305 is further operably coupled to a data store 345. The data store 345 includes the note classifier 120, the LLM 125, the predefined prompt 140, the user profile 145, the association data store 235, and prompt selection rules 350. For example, upon receiving the multimedia content 105 from the user 110 (e.g., one of the multiple users 205). For example, the NGE 115 may classify the multimedia content 105. For example, the NGE 115 may apply the LLM 125 to the multimedia content 105 based on the classification and the user profile 145. In some implementations, the NGE 115 may select the predefined prompt 140 based on the prompt selection rules 350.

[0043] For example, the UPLE 330 may learn user preference based on user interactions (e.g., received user instruction prompts) with the transformed content 135. For example, the UPLE 330 may update the prompt selection rules 350 based on the learned user behavior.Docket #: 1040-05WO / US

[0044] For example, after NGE 115 generated the transformed content 135, the INDE 230 may deliver the transformed content 135 to a next user based on information stored in the association data store 235. For example, the INDE 230 may, upon receiving registration notes from a nurse, deliver an edited note to a doctor scheduled to see a patient related to the registration notes based on the association data store 235. For example, the data integration engine 220 may, based on the association data store 235, integrate more than one asynchronously generated clinical notes into the integrated content 215.

[0045] FIG. 4 is a flowchart illustrating an exemplary interactive electronic medical record generation method 400. For example, the method 400 may be performed by the NGE 115. For example, the method 400 may be formed by the ACGS 300. In this example, the method 400 begins in step 405 when content of a user is received. For example, the NGE 115 may receive multimedia content 105 from the user 110 (e.g., a medical assistant or doctor) to initiate a note generation (e.g., after a clinical note is written after an interview with a patient).

[0046] In step 410, the content received from the user is classified. For example, the note classifier 120 may analyze the multimedia content 105 to categorize it. For example, the note classifier 120 may classify the multimedia content 105 to be a consultation note. For example, the note classifier 120 may classify the multimedia content 105 to be an after-visit summary. For example, the note classifier 120 may classify the multimedia content 105 to be a medical image analyzing report. For example, the note classifier 120 may classify the multimedia content 105 to be a blood test result report.

[0047] In step 415, a user profile associated with the user is retrieved. For example, the NGE 115 may retrieve the user profile 145 to access user-specific preferences and / or settings. In step 420, predefined prompt(s) are selected based on the user profile. For example, the predefined prompt 140 may be selected by the note editing LLM 125 based on the user profile 145 and the classification determined by the note classifier 120.

[0048] After selection, the selected predefined prompt(s) and the content are applied to a note editing LLM to generate an edited note in step 425. For example, the note editing LLM 125 may apply the predefined prompt 140 and the multimedia content 105 to generate a refined clinical based on user preference associated with the user profile 145. In step 430, the edited note is transmitted to a user device of the user. For example, the NGE 115 may send the transformed content 135 (e.g., the completed clinical note) to the user device 130 for the user 110 to review.Docket #: 1040-05WO / US

[0049] At a decision point 435, it is determined whether user feedback is received. For example, the user device 130 may transmit feedback when modifications are needed on the edited note (e.g., using an in-app chatbot). If no user feedback is received, the method 400 ends. In step 440, if user feedback is received, an updated edited note is transmitted by applying the feedback using the note editing LLM. For example, the NGE 115 may apply adjustments to the transformed content 135 based on user comment received from the user device 130.

[0050] At a decision point 445, it is determined if more user feedback is received. For example, the system may check if any additional comment is received from the user device 130. If more user feedback is received, the step 440 is repeated. If no more user feedback is received, at a decision point 450, it is determined whether to update the user profile, the predefined prompts, and / or the classifier. For example, the UPLM 150 may prompt the user 110 whether to adjust user preferences or refine the classifier based on recent feedback trends and / or repeated modifications.

[0051] In step 455, if an update is to be performed, the user profile, predefined prompts, and / or classifier are updated. For example, the UPLM 150 may adjust the user profile 145, predefined prompt 140, or the note classifier 120 to streamline future interactions based on observed preferences and classifications. If no update is to be performed, the method 400 ends.

[0052] FIG. 5 is a flowchart illustrating an exemplary continuous medical record integration method. For example, the method 500 may be performed by the CCGS 200. In this example, the method 500 begins in step 505 when an asynchronous note is received. For example, the data integration engine 220 may receive a note generated by one of the multiple users 205. For example, the multiple users 205 may include a doctor. For example, the multiple users 205 may include a medical assistant. For example, the note may contribute to a patient's record at different times.

[0053] In step 510, an association between the asynchronous note and other stored notes is identified. For example, the data integration engine 220 may determine whether the received note is linked to any existing notes. For example, the existing notes may be related to the same patient. For example, the existing notes may be related to the same visit. In step 515, the note is added to the associated stored note. For example, the data integration engine 220 may append the asynchronous note to an ongoing record for a patient. For example, appending may combine contributions from multiple users into a single, coherent document. In someDocket #: 1040-05WO / USimplementations, the data integration engine 220 may insert relevant new information into the stored note (e.g., an EHR).

[0054] At a decision point 520, it is determined whether the stored note is associated with another user. For example, the INDE 230 may check if the stored note needs to be delivered to a different user (e.g., a doctor who will see the patient next). If the stored note is associated with another user, in step 525, the stored note is edited based on a user profile of the associated user. For example, the note editing LLM 125 may adjust the language in the note. For example, the note editing LLM 125 may adjust details in the note. For example, adjusting may align the note with the preferences of the receiving user (e.g., as described with reference to FIGS. 1 and 4). For example, adjusting may align the note with the preferences of the receiving user (e.g., as described with reference to FIGS. 1 and 4). In step 530, the stored note is transmitted to the associated user identified in step 525. For example, the INDE 230 may deliver the integrated content 215 to one or more of the multiple users 205. After the step 530 or if no associated is determined at the decision point 520, the edited note is stored in step 535, and the method 500 ends. For example, the updated version of the stored note may be stored in the patient history database 240.

[0055] FIG. 6 shows an example predefined prompt training system. In this example, a predefined prompt training system (PPTS 600) may be configured adaptively train the predefined prompt 140 based on the user profile 145 and the user-initiated feedback 155. As shown, the PPTS 600 includes the UPLM 150. The UPLM 150 may, for example, be configured to generate an updated predefined prompt 615 based on a prompt training data package (PTDP 605).

[0056] In the depicted example, the PTDP 605 includes the user-initiated feedback 155, the user profile 145, and historical user behavior 610. For example, the historical user behavior 610 may include previous editing operations, frequency of content modifications, and response patterns to system-generated suggestions. In some implementations, the UPLM 150 may correlate the historical user behavior 610 with the record type 630 and the transformed content type 635 to identify a pattern of preferred prompt behaviors. The UPLM 150 may generate the updated predefined prompt 615 based on the PTDP 605 and store the updated predefined prompt 615 in the predefined prompt 140 repository for later use. For example, the updated predefined prompt 615 may update the prompt selection rules 350. For example, the updated predefined prompt 615 may update the predefined prompt 140.Docket #: 1040-05WO / US

[0057] For example, the PTDP 605 may aggregate the user-initiated feedback 155 with the user profile 145 and the historical user behavior 610 to generate training features that represent a user’s editing intent and prompt preference. The PTDP 605 may further normalize and weigh these features based on the record type 630 so that the updated predefined prompt 615 can adapt to different document types or content contexts (e.g., progress note versus discharge summary).

[0058] For example, the historical user behavior 610 may include a trend of a particular user in revising the transformed content 135. For example, a user may regularly insert clarifying details for diagnostic sections or remove redundant phrases from summaries. The UPLM 150 may use such patterns to generate a bias parameter for future prompt selection so that the system proactively includes or excludes certain content elements during transformation. For example, the updated predefined prompt 615 may include a new prompt. For example, the updated predefined prompt 615 may include an instruction to refine one or more existing prompts.

[0059] In this example, the user-initiated feedback 155 includes user-generated instructions 620 and user-defined rules 625. For example, the user-generated instructions 620 may specify that a particular section of a note be expanded with additional context or that specific terminology be replaced with more patient-friendly language. The UPLM 150 may analyze the user-generated instructions 620 together with the historical user behavior 610 to identify repeated patterns that inform future prompt selection and refinement. For example, the user-defined rules 625 may include user instructions in format, terminology, phrasing, or other directive integrating content for a user.

[0060] As shown, the PPTS 600 includes a record type 630. In some implementations, the user-defined rules 625 may include conditional logic linking a record type 630 with preferred actions (e.g., “always include dosage field for prescription records,” “omit patient ID in summary section”). As an illustrative example without limitation, the PPTS 600 may continuously learn from each user’s feedback to update the predefined prompt 140 in real time or near real-time.

[0061] The UPLM 150 may receive a transformed content type 635. For example, the UPLM 150 may analyze the user-generated instructions 620 in combination with the record type 630 to derive prompt adjustments amongst the transformed content type 635.

[0062] As an illustrative example without limitation, in operation, the PPTS 600 may receive the user-initiated feedback 155 and the user profile 145 to generate the updated predefinedDocket #: 1040-05WO / USprompt 615 based on the record type 630 and the transformed content type 635. For example, when a user repeatedly modifies clinical notes to expand diagnostic details in consultation records, the UPLM 150 may adjust the predefined prompt 140 to automatically request additional findings when similar record types are processed in the future. In some examples, when a user frequently reformats discharge summaries to simplify terminology for patientfacing communication, the updated predefined prompt 615 may instruct the system to generate content with simplified phrasing for the transformed content type 635. Various embodiments may advantageously enable real-time or near real-time adaptation of prompts across different record types and content formats, improving system responsiveness, personalization, and overall computational efficiency without retraining the LLM 125.

[0063] FIG. 7 shows an example association data store. In this example, the association data store 235 may include multiple types of relational and contextual information used by the INDE 230 and the data integration engine 220 to determine workflow context, record association, and target destinations of integrated content 215. As shown, the association data store 235 includes user relationships 705, record types 720, workflow information 725, and association rules 730.

[0064] In the depicted example, the user relationships 705 may define how users interact with one another within a collaborative environment. For example, the user relationships 705 may include internal relationships 710. The internal relationships 710 may include relationships between members of the same clinical team. For example, the user relationships 705 may include external relationships 715. For example, the external relationships 715 may include connections between external organizations. For example, the external relationships 715 may include external specialists. For example, the internal relationships 710 may include an assignment between a nurse and a supervising doctor. In some examples, the external relationships 715 may include a referral from a primary care physician to a specialist.

[0065] The record types 720 may include classifications of the data input 210 (e.g., registration notes, diagnostic reports, treatment plans, imaging studies, laboratory results, prescription records). For example, the record types 720 may include classifications of the multimedia content 105.

[0066] The workflow information 725 may include procedural sequences associated with the record types 720 and the user relationships 705. The workflow information 725 may include temporal sequences defining order or timing of actions. The workflow information 725 may include scheduling information. The workflow information 725 may include patient statusDocket #: 1040-05WO / USindicators. The workflow information 725 may include pending action flags. For example, the procedural and temporal sequences may be used to maintain the order of collaboration among users. For example, the scheduling information may be used to coordinate asynchronous updates between different workflow stages. For example, the patient status indicators may track progress of a patient’s record through different workflow stages. For example, the pending action flags may identify items for follow-up or confirmation before a record can advance. Various embodiments may advantageously improve traceability and / or facilitate continuity of record updates throughout an asynchronous workflow.

[0067] The association rules 730 may include logical relationships that link the user relationships 705, the record types 720, and the workflow information 725. As an illustrative example, the association rules 730 may specify that a note categorized as a “lab result” is to be routed to a physician assigned to a specific patient or department. The association rules 730 may (e.g., also) define conditional delivery behaviors (e.g., “if test results exceed a defined threshold, alert the attending doctor immediately”), for example.

[0068] As an illustrative example without limitation, the INDE 230 may access the association data store 235 to automatically determine the most appropriate recipient and transformation content for the data input 210. For example, when a registration note from a nurse includes an entry for a specific patient, the association data store 235 may identify a corresponding physician through the internal relationships 710 and the workflow information 725. Various embodiments may advantageously reduce manual routing errors and improve efficiency.

[0069] Although various embodiments have been described with reference to the figures, other embodiments are possible.

[0070] For example, although referred to in some embodiments as a “note editing LLM 125”, LLMs (e.g., LLM 125) may include one or more language model architectures capable of processing natural language inputs and generating contextually appropriate outputs. For example, the LLM 125 may be based on transformer architectures, such as GPT (Generative Pre-trained Transformer) models. The LLM 125 may include BERT (Bidirectional Encoder Representations from Transformers) models. The LLM 125 may, for example, include T5 (Text-to-Text Transfer Transformer) models. The LLM 125 may, for example, include other neural network architectures designed for natural language processing tasks. The LLM 125 may, for example, include billions of parameters trained on diverse text corpora to understand and generate target-type outputs (e.g., text, code, other media) across multiple domains and languages. In some implementations, the LLM 125 may be a general-purpose language modelDocket #: 1040-05WO / US(e.g., GPT-4, CLAUDE, LLAMA) that has been pre-trained on broad datasets encompassing medical literature, clinical documentation, general knowledge, and / or conversational text. In some implementations, the LLM 125 may be a domain-specific model that has been pretrained and / or fine-tuned on specialized corpora (e.g., medical journals, clinical notes, pharmaceutical documentation, healthcare regulatory materials). In some embodiments, an LLM 125 may be a small language model (SLM) and / or combination of small and large language models. In some embodiments, the LLM 125 may include one or more models other than language models.

[0071] In some embodiments, the LLM 125 may, by way of example and not limitation, be configured as a note editing LLM. In some embodiments, the LLM 125 may, for example, be configured to perform various natural language processing tasks other than or in addition to note generation and / or editing. For example, the LLM 125 may perform text summarization to condense lengthy clinical documentation into concise summaries. The LLM 125 may, for example, perform named entity recognition to identify and extract specific information such as patient names, medication names, dosages, dates, diagnoses, and procedure codes from unstructured text. The LLM 125 may, for example, perform sentiment analysis, such as to assess tone or urgency in clinical communications. The LLM 125 may, for example, perform language translation, such as to convert clinical notes between different languages for multilingual healthcare environments. The LLM 125 may, for example, perform question answering to respond to specific queries about patient history or treatment protocols based on available documentation. The LLM 125 may, for example, perform text classification to categorize clinical notes into appropriate record types, diagnostic categories, and / or urgency levels. The LLM 125 may, for example, perform information extraction to identify relationships between clinical entities such as symptoms, diagnoses, treatments, and outcomes. The LLM 125 may, for example, be configured as a report generation model(s). The LLM 125 may, for example, be configured to generate multimedia. Various embodiments may advantageously leverage the broad capabilities of the LLM 125 across multiple natural language processing tasks while maintaining computational efficiency through prompt-based adaptation rather than model retraining.

[0072] In some implementations, the LLM 125 may be deployed as a cloud-based service accessible via application programming interfaces (APIs). Such embodiments may, for example, advantageously enable the ACGS 100 to leverage powerful language models without demanding local computational resources for model hosting. For example, the LLM 125 mayDocket #: 1040-05WO / USbe accessed through API calls to services such as commercially available and / or open-source language model services.

[0073] In some implementations, the LLM 125 may, for example, be deployed locally, such as on dedicated hardware infrastructure (e.g., GPU clusters, tensor processing units). Local embodiments may, for example, advantageously maintain data privacy, reduce latency, and / or facilitate compliance with regulatory guidelines such as HIPAA. For example, in healthcare environments handling protected health information (PHI), the LLM 125 may be deployed within secure, on-premises infrastructure, such as to prevent transmission of sensitive patient data to external services.

[0074] The LLM 125 may, for example, be containerized using technologies such as Docker or Kubernetes. Containerized technologies may, for example, advantageously enable scalable deployment across distributed computing environments. The LLM 125 may, for example, be adjusted for inference efficiency. Such embodiments may, for example, be adjusted through techniques such as model quantization, pruning, and / or distillation. Adaptation may, for example, reduce computational demands while maintaining acceptable performance levels. Various embodiments may advantageously provide flexible deployment options for the LLM 125 to balance performance, cost, privacy, and regulatory compliance across different use cases and / or organizational contexts.

[0075] In various embodiments, the user profile 145 may be implemented as a structured data object containing user-specific information for prompt selection and content transformation. For example, a user profile data structure may include a user identifier (e.g., "USER 12345"), user name (e.g., "Dr. Jane Smith"), user role (e.g., "Physician"), department (e.g., "Cardiology"), creation date, and last update timestamp. The user profile 145 may, for example, include user-specific weightings such as detail level (e.g., 0.85), formality (e.g., 0.72), technical terminology preference (e.g., 0.90), patient-friendly language preference (e.g., 0.35), differential diagnosis inclusion preference (e.g., 0.88), and treatment rationale inclusion preference (e.g., 0.75). These weightings may, for example, be represented as numerical values (e.g., floating-point numbers between 0 and 1) that quantify user preferences for various content characteristics.

[0076] The user profile 145 may, for example, include record type preferences. For example, for consultation notes, the user profile 145 may store a preferred prompt identifier (e.g., "PROMPT_CN_003"), custom sections (e.g., "detailed_history," "assessment," "plan"), sections to automatically expand (e.g., "assessment," "differential diagnosis"), and sections toDocket #: 1040-05WO / USomit (e.g., "billing codes"). For after-visit summaries, the user profile 145 may, for example, store a different preferred prompt identifier (e.g., "PROMPT AVS 012"), different custom sections (e.g., "visit_summary," "medications," "follow_up"), sections to automatically expand (e.g., "follow_up"), and a flag indicating whether to simplify terminology (e.g., true). For discharge summaries, the user profile 145 may, for example, store yet another preferred prompt identifier (e.g., "PROMPT DS 007"), custom sections (e.g., "hospital course," "discharge medications," "follow up care"), and sections to automatically expand (e.g., "hospital course," "complications").

[0077] The user profile 145 may, for example, include terminology preferences. For example, the user profile 145 may store preferred terms such as mappings from technical terms to simplified terms (e.g., "myocardial infarction" to "heart attack," "hypertension" to "high blood pressure"). The user profile 145 may, for example, store terms to avoid (e.g., "expired," "negative outcome"). The user profile 145 may, for example, include formatting preferences such as date format (e.g., "MM / DD / YYYY"), time format (e.g., "12-hour"), measurement units (e.g., "imperial"), and list style (e.g., "bulleted").

[0078] The user profile 145 may, for example, include a feedback history summary. For example, the feedback history summary may include total feedback count (e.g., 247), last feedback date, and common modifications. The common modifications may, for example, include modification type (e.g., "expand_section"), target section (e.g., "assessment"), and frequency (e.g., 45 occurrences). Other common modifications may, for example, include adding detail to treatment plans (e.g., 38 occurrences) or simplifying language in patient instructions (e.g., 22 occurrences).

[0079] In various embodiments, the user-initiated feedback 155 may be captured and stored in a structured format for processing by the UPLM 150. For example, a feedback data structure may include a feedback identifier (e.g., "FB_98765"), user identifier (e.g., "USER_12345"), timestamp (e.g., "2024-1 l-26T14:22:00Z"), content identifier (e.g., "CONTENT_54321"), record type (e.g., "consultation note"), original prompt identifier (e.g., "PROMPT CN 003"), and feedback type (e.g., "modification").

[0080] The user-initiated feedback 155 may, for example, include user-generated instructions. For example, the user-generated instructions 620 may include instruction text (e.g., "Add more detail about the patient's cardiac history and include specific ejection fraction measurements"), instruction type (e.g., "expand_content"), target section (e.g., "history _of_present_illness"), and priority (e.g., "high").Docket #: 1040-05WO / US

[0081] The user-initiated feedback 155 may, for example, include user-defined rules 625. For example, a user-defined rule may include a rule identifier (e.g., "RULE 001"), rule text (e.g., "Always include ejection fraction when discussing heart failure patients"), condition (e.g., record type is "consultation note" and diagnosis contains "heart failure" or "cardiomyopathy"), and action (e.g., include field "ejection fraction" in section "assessment").

[0082] The user-initiated feedback 155 may, for example, include editing operations. For example, an editing operation may include an operation identifier (e.g., "EDIT 001"), operation type (e.g., "insert_text"), location (e.g., section "hi story _of_present_illness," paragraph 2, position "after_sentence_3"), inserted text (e.g., "Patient's most recent echocardiogram from 10 / 15 / 2024 showed an ejection fraction of 35%, consistent with moderate systolic dysfunction"), and timestamp. Another editing operation may, for example, include operation type "modify text," location (e.g., section "assessment," paragraph 1, sentence 2), original text (e.g., "Patient has heart failure"), modified text (e.g., "Patient has heart failure with reduced ejection fraction (HFrEF), NYHA Class II"), and timestamp.

[0083] The user-initiated feedback 155 may, for example, include evaluation instructions. For example, evaluation instructions may include satisfaction rating (e.g., 3 on a scale of 5), indication of whether improvement is needed (e.g., true), specific issues (e.g., "Insufficient detail in cardiac assessment," "Missing quantitative measurements"), and positive aspects (e.g., "Good organization of sections," "Appropriate medication list").

[0084] The user-initiated feedback 155 may, for example, include context metadata. For example, context metadata may include session identifier (e.g., "SESSION_789"), device type (e.g., "desktop"), time to first edit (e.g., 45 seconds), total editing time (e.g., 180 seconds), and number of edits (e.g., 2).

[0085] In various embodiments, the prompt-adjustment parameters generated by the UPLM 150 may be structured to capture learned preferences and modifications. For example, a prompt-adjustment parameters data structure may include an adjustment identifier (e.g., "ADJ_11223"), user identifier (e.g., "USER_12345"), generation timestamp (e.g., "2024-11-26T14:25:00Z"), source feedback identifiers (e.g., "FB_98765," "FB_98764," "FB_98763"), record type (e.g., "consultation note"), and adjustment type (e.g., "prompt modification").

[0086] The prompt-adjustment parameters may, for example, include weight adjustments. For example, a weight adjustment for detail level may include previous value (e.g., 0.85), adjustment delta (e.g., 0.05), new value (e.g., 0.90), and confidence score (e.g., 0.87). AnotherDocket #: 1040-05WO / USweight adjustment for including quantitative data may, for example, include previous value (e.g., 0.65), adjustment delta (e.g., 0.15), new value (e.g., 0.80), and confidence score (e.g., 0.92).

[0087] The prompt-adjustment parameters may, for example, include prompt modifications. For example, a prompt modification may include modification type (e.g., "add instruction"), target prompt identifier (e.g., "PROMPT_CN_003"), instruction text (e.g., "When discussing cardiac conditions, always include relevant quantitative measurements such as ejection fraction, troponin levels, or BNP values when available in the source content"), priority (e.g., "high"), and applicable conditions (e.g., diagnosis categories include "cardiovascular" and "heart failure," record types include "consultation note" and "progress note"). Another prompt modification may, for example, include modification type "modify section template," target section (e.g., "assessment"), and template changes (e.g., add fields "ejection_fraction" and "nyha_class," field format "structured").

[0088] The prompt-adjustment parameters may, for example, include rule updates. For example, a rule update may include rule identifier (e.g., "RULE 001"), rule action (e.g., "create"), and rule definition including rule text, condition, and action as described above.

[0089] The prompt-adjustment parameters may, for example, include learning metadata. For example, learning metadata may include pattern confidence (e.g., 0.89), sample size (e.g., 3), consistency score (e.g., 0.91), and / or recommendation (e.g., "apply_immediately").

[0090] In various embodiments, the historical user behavior 610 may aggregate patterns over time to inform the UPLM 150. For example, a historical user behavior data structure may include user identifier (e.g., "USER 12345"), behavior period with start date and end date, total content generated (e.g., 1247), total feedback events (e.g., 247), feedback rate (e.g., 0.198), and record type distribution (e.g., consultation notes: 456, progress notes: 312, aftervisit summaries: 289, discharge summaries: 190).

[0091] The historical user behavior 610 may, for example, include editing patterns. For example, an editing pattern may include pattern identifier (e.g., "PATTERN 001"), pattern type (e.g., "section_expansion"), target section (e.g., "assessment"), frequency (e.g., 45), average expansion length (e.g., 127 characters), common additions (e.g., "quantitative measurements," "differential diagnosis," "clinical reasoning"), and applicable record types (e.g., "consultation note," "progress note"). Another editing pattern may, for example, include pattern type "terminology simplification," target section (e.g., "patient_instructions"), frequency (e.g., 22), common replacements (e.g., "myocardialDocket #: 1040-05WO / USinfarction" to "heart attack," "hypertension" to "high blood pressure," "anti coagulation" to "blood thinner"), and applicable record types (e.g., "after visit summary," "discharge summary"). Yet another editing pattern may, for example, include pattern type "add_missing_field," target fields (e.g., "ejection_fraction," "nyha_class," "medication_dosages"), frequency (e.g., 38), and applicable record types (e.g., "consultation note").

[0092] The historical user behavior 610 may, for example, include temporal patterns. For example, temporal patterns may include time-of-day preferences (e.g., morning: detail level 0.92, formality 0.75; afternoon: detail level 0.85, formality 0.72; evening: detail level 0.78, formality 0.68) and day-of-week patterns (e.g., Monday: average edits per note 2.3; Friday: average edits per note 1.8).

[0093] The historical user behavior 610 may, for example, include response patterns. For example, response patterns may include average time to first edit (e.g., 52 seconds), average total editing time (e.g., 165 seconds), average edits per content (e.g., 1.9), and confirmation acceptance rate (e.g., 0.87).

[0094] The historical user behavior 610 may, for example, include satisfaction metrics. For example, satisfaction metrics may include average satisfaction rating (e.g., 4.2 on a scale of 5), satisfaction trend (e.g., "improving"), and trend slope (e.g., 0.15).

[0095] In various embodiments, the prompt selection rules 350 may be stored in a structured format that enables dynamic selection based on context. For example, a prompt selection rules data structure may include rule set identifier (e.g., "RULESET USER 12345"), user identifier (e.g., "USER_12345"), last updated timestamp, and version (e.g., "3.7").

[0096] The prompt selection rules 350 may, for example, include selection rules. For example, a selection rule may include rule identifier (e.g., "SEL RULE OOl"), priority (e.g., 1), conditions (e.g., record type is "consultation note," user role is "Physician," department is "Cardiology," content contains "heart failure" or "cardiomyopathy"), action (e.g., select prompt "PROMPT_CN_003_CARDIAC_ENHANCED," apply modifiers "include quantitative measurements, " "expand assessment section, " "include ejection fraction"), and confidence (e.g., 0.92). Another selection rule may, for example, include priority 2, conditions (e.g., record type is "after_visit_summary," patient age greater than 65, content complexity is "high"), action (e.g., select prompt "PROMPT AVS 012 SIMPLIFIED," apply modifiers "simplify terminology," "add visual aids," "expand follow up instructions"), and confidence (e.g., 0.85). YetDocket #: 1040-05WO / USanother selection rule may, for example, include priority 3, conditions (e.g., record type is "discharge summary," length of stay greater than 7 days, complications present is true), action (e.g., select prompt "PROMPT DS 007 DETAILED," apply modifiers "expand hospital course," "detail complications," "comprehensive medication reconciliation"), and confidence (e.g., 0.88).

[0097] The prompt selection rules 350 may, for example, include default rules. For example, default rules may specify default prompts for each record type (e.g., consultation note: "PROMPT CN OOl," progress note: "PROMPT PN OOl," after-visit summary: "PROMPT AVS OOl," discharge summary: "PROMPT DS OOl").

[0098] The prompt selection rules 350 may, for example, include rule application statistics. For example, rule application statistics may include total applications (e.g., 1247), successful applications (e.g., 1189), success rate (e.g., 0.953), and average confidence (e.g., 0.87).

[0099] In various embodiments, the PTDP 605 may aggregate multiple data sources for training. For example, a PTDP data structure may include PTDP identifier (e.g., "PTDP 67890"), user identifier (e.g., "USER 12345"), creation timestamp, and training purpose (e.g., "update_prompt_for_consultation_notes").

[0100] The PTDP 605 may, for example, include user-initiated feedback information. For example, the user-initiated feedback information may include feedback identifiers (e.g., "FB_98765," "FB_98764," "FB_98763"), aggregated instructions (e.g., instruction category "expand content," frequency 3, common targets "assessment" and "hi story _of_present_illness," common additions "quantitative measurements" and "clinical_reasoning"), and aggregated rules (e.g., rule category "include_field," frequency 3, common fields "ejection fraction" and "nyha class").

[0101] The PTDP 605 may, for example, include user profile information. For example, the user profile information may include user-specific weightings (e.g., detail level 0.90, technical terminology 0.90, include quantitative data 0.80) and record type preferences (e.g., for consultation notes, auto-expand sections "assessment" and "differential diagnosis").

[0102] The PTDP 605 may, for example, include historical user behavior information. For example, the historical user behavior information may include relevant patterns (e.g., pattern "PATTERN_001" with type "section_expansion," target section "assessment," frequency 45; pattern "PATTERN 003 " with type "add_missing_field," target fields "ejection_fraction," frequency 38), satisfaction trend (e.g., "improving"), and average satisfaction rating (e.g., 4.2).Docket #: 1040-05WO / US

[0103] The PTDP 605 may, for example, include record type (e.g., "consultation note") and transformed content type (e.g., "clinical note detailed").

[0104] The PTDP 605 may, for example, include training features. For example, training features may include feature vector (e.g., [0.90, 0.90, 0.80, 0.75, 0.88, 0.85]), feature names (e.g., "detail_level," "technical_terminology," "include_quantitative_data," "formality," "include differential diagnosis," "include treatment rationale"), and feature weights (e.g., [0.25, 0.15, 0.20, 0.10, 0.15, 0.15]).

[0105] The PTDP 605 may, for example, include context metadata. For example, context metadata may include department (e.g., "Cardiology"), user role (e.g., "Physician"), experience level (e.g., "senior"), and typical patient complexity (e.g., "high").

[0106] In various embodiments, the updated predefined prompt 615 may be stored with versioning and metadata. For example, an updated predefined prompt data structure may include prompt identifier (e.g., "PROMPT_CN_003_CARDIAC_ENHANCED"), base prompt identifier (e.g., "PROMPT_CN_003"), version (e.g., "2.1"), creation timestamp, user identifier (e.g., "USER_12345"), and record type (e.g., "consultation_note").

[0107] The updated predefined prompt 615 may, for example, include prompt text. For example, the prompt text may be: "Generate a detailed consultation note based on the provided multimedia content. Structure the note with the following sections: Chief Complaint, History of Present Illness, Past Medical History, Medications, Allergies, Physical Examination, Assessment, and Plan. In the Assessment section, provide a comprehensive analysis including differential diagnosis and clinical reasoning. When discussing cardiac conditions, always include relevant quantitative measurements such as ejection fraction, troponin levels, or BNP values when available in the source content. Use technical medical terminology appropriate for physician-to-physician communication. Make all sections thoroughly detailed with specific clinical findings and measurements."

[0108] The updated predefined prompt 615 may, for example, include prompt modifiers. For example, a prompt modifier may include modifier identifier (e.g., "MOD 001 "), modifier type (e.g., "section_enhancement"), target section (e.g., "assessment"), and instruction (e.g., "Expand the assessment section with differential diagnosis and clinical reasoning. Include quantitative measurements when discussing cardiac conditions"). Another prompt modifier may, for example, include modifier identifier (e.g., "MOD_002"), modifier type (e.g., "included field"), condition (e.g., diagnosis contains "heart failure" or "cardiomyopathy"),Docket #: 1040-05WO / USand instruction (e.g., "Ensure ejection fraction and NYHA class are included when discussing heart failure").

[0109] The updated predefined prompt 615 may, for example, include parameters. For example, parameters may include maximum length (e.g., 2000 characters), temperature (e.g., 0.7), detail level (e.g., "high"), formality (e.g., "professional"), include differential diagnosis flag (e.g., true), and / or include quantitative data flag (e.g., true).

[0110] The updated predefined prompt 615 may, for example, include applicable conditions. For example, applicable conditions may include user roles (e.g., "Physician," "Physician Assistant"), departments (e.g., "Cardiology," "Internal Medicine"), and record types (e.g., "consultation note," "progress note").

[0111] The updated predefined prompt 615 may, for example, include performance metrics. For example, performance metrics may include usage count (e.g., 156), average satisfaction rating (e.g., 4.5), edit frequency (e.g., 0.12), and average edits peruse (e.g., 0.8).

[0112] The updated predefined prompt 615 may, for example, include parent prompt identifier (e.g., "PROMPT CN 003") and derivation metadata. For example, derivation metadata may include derived from feedback count (e.g., 3), learning confidence (e.g., 0.89), and last training date.

[0113] Illustrative examples of data structures are provided, by way of example and not limitation, below.

[0114] The user profile 145 may be implemented as a structured data object containing userspecific information for prompt selection and content transformation. An example user profile data structure is depicted below by way of illustration:{"user id" : "USER 12345" ,"user name" : "Dr . Jane Smith" ,"user role" : "Physician" ,"department" : "Cardiology" ,"created date" : "2024-01-15T08 : 30 : 00Z" ,"last updated" : "2024-11-26T14 : 22 : 00Z" ,"user specific weightings" : {"detail level" : 0. 85,"formality" : 0.72 ,"technical terminology" : 0. 90,"patient friendly language" : 0.35,"include differential diagnosis" : 0. 88 ,"include treatment rationale" : 0.75} ,"record type preferences" : {"consultation note" : {"preferred prompt id" : "PROMPT CN 003" ,"custom sections" : ["detailed history" , "assessment" , "plan" ] , "auto expand sections" : ["assessment" , "differential diagnosis" ] ,Docket #: 1040-05WO / US"omit sections" : [ "billing codes" ]} ,"after visit summary" : {"preferred prompt id" : " PROMPT AVS 012" ," custom sections" : [ "visit summary" , "medications" , " follow up" ] , "auto expand sections" : [ " follow up" ] ," simpli fy terminology" : true} ,"dis charge summary" : {"preferred prompt id" : " PROMPT DS 007" ," custom sections" : [ "hospital course" , "dis charge medications" , " follow up care" ] ,"auto expand sections" : [ "hospital course" , " complications" ]}} ,"terminology preferences" : {"preferred terms" : {"myocardial infarction" : "heart attack" ,"hypertension" : "high blood pres sure"} ,"avoid terms" : [ "expired" , "negative outcome" ]} ," formatting preferences" : {"date_format" : "MM / DD / YYYY" ,"time format" : " 12-hour" ,"measurement units" : "imperial" ,"list style" : "bulleted"} ," feedback history summary" : {"total feedback count" : 247 ,"last feedback date" : "2024- 11-26T14 : 22 : 00Z" ," common modi fications" : [{ "type" : "expand section" , " section" : "as ses sment" , " frequency" : 45 } , { "type" : "add detail" , " section" : "treatment plan" , " frequency" : 38 } , { "type" : " simpli fy language" , " section" : "patient instructions" , " frequency" : 22 }]}}

[0115] A user-initiated feedback 155 may, for example, be captured and stored in a structured format, such as for processing by the UPLM 150. An example feedback data structure is depicted below by way of illustration:{" feedback id" : " FB 98765" ,"user id" : "USER 12345" ,"time st amp" : "2024- l l-26T14 : 22 : 00Z" ," content_id" : "CONTENT_54321" ," record type" : " consultation note" ,"original prompt id" : " PROMPT ON 003" ," feedback type" : "modi fication" ,"user generated instructions" : {"instruction text" : "Add more detail about the patient ' s cardiac history and include speci fic ej ection fraction measurements" ,"instruction type" : "expand content" ,"target section" : "history of present illnes s" ,"priority" : "high"} ,Docket #: 1040-05WO / US"user defined rules" : [{" rule_id" : "RULE_001" ," rule text" : "Always include ej ection fraction when dis cus sing heart failure patients" ," condition" : {" record type" : " consultation note" ,"diagnosis contains" : [ "heart failure" , " cardiomyopathy" ] } ,"action" : {"type" : "include field" ," field name" : "ej ection fraction" ," section" : "as ses sment"}}] ,"editing operations" : [{"operation id" : "EDIT 001" ,"operation type" : "insert text" ,"location" : {" section" : "history of present illnes s" ,"paragraph" : 2 ,"position" : "after sentence 3"} ,"inserted text" : " Patient ' s most recent echocardiogram from 10 / 15 / 2024 showed an ej ection fraction of 35% , consistent with moderate systolic dys function . " ,"time st amp" : "2024- l l-26T14 : 22 : 15Z"} ,{"operation id" : "EDIT 002" ,"operation type" : "modi fy text" ,"location" : {" section" : "as ses sment" ,"paragraph" : 1 ," sentence" : 2} ,"original text" : " Patient has heart failure . " ,"modi fied text" : " Patient has heart failure with reduced ej ection fraction ( HFrEF?, NYHA Clas s I I . " ,"time st amp" : " 2024- 11-26T14 : 23 : 30Z"}] ,"evaluation instructions" : {" satis faction rating" : 3 ," satis faction s cale" : 5 ,"improvement needed" : true ," speci fic is sues" : [" Insufficient detail in cardiac as ses sment" ,"Mis sing quantitative measurements"] ,"positive aspects" : ["Good organi zation of sections" ,"Appropriate medication list"]} ," context metadata" : {"session_id" : " SESSION_789" ,"device type" : "des ktop" ,Docket #: 1040-05WO / US"time to first edit" : 45 ,"total editing time" : 180 ,"number of edits" : 2}}

[0116] Prompt-adjustment parameters generated by the UPLM 150 may be structured, for example, to capture learned preferences and / or modifications. An example prompt-adjustment parameter data structure is depicted below by way of illustration:{"adj ustment id" : "ADJ 11223" ,"user id" : "USER 12345" ,"generated timestamp" : "2024- 11-26T14 : 25 : 00Z" ," source feedback ids" : [ " FB 98765" , " FB 98764" , " FB 98763" ] ," record type" : " consultation note" ,"adj ustment type" : "prompt modi fication" ,"weight adj ustments" : {"detail level" : {"previous value" : 0 . 85 ,"adj ustment delta" : 0 . 05 ,"new value" : 0 . 90 ," confidence s core" : 0 . 87} ,"include quantitative data" : {"previous value" : 0 . 65 ,"adj ustment delta" : 0 . 15 ,"new value" : 0 . 80 ," confidence s core" : 0 . 92}} ,"prompt modi fications" : [{"modi fication type" : "add instruction" ,"target prompt id" : " PROMPT ON 003" ,"instruction text" : "When dis cus sing cardiac conditions , always include relevant quantitative measurements such as ej ection fraction, troponin levels , or BNP values when available in the source content . " , "priority" : "high" ,"applicable conditions" : {"diagnosis categories" : [ " cardiovas cular" , "heart failure" ] , " record types" : [ " consultation note" , "progres s note" ]}} ,{"modi fication type" : "modi fy section template" ,"target section" : "as ses sment" ,"template changes" : {"add included fields" : [ "ej ection fraction" , "nyha clas s" ] , " field format" : " structured"}}] ," rule updates" : [{" rule_id" : "RULE_001" ," rule action" : " create" ," rule definition" : {Docket #: 1040-05WO / US" rule text" : "Always include ej ection fraction when dis cus sing heart failure patients" ," condition" : {" record type" : " consultation note" ,"diagnosis contains" : [ "heart failure" , " cardiomyopathy" ]} ,"action" : {"type" : "include field" ," field name" : "ej ection fraction" ," section" : "as ses sment"}}}] ,"learning metadata" : {"pattern confidence" : 0 . 89 ," sample si ze" : 3 ," consistency s core" : 0 . 91 ," recommendation" : "apply immediately"}}

[0117] A historical user behavior may, for example, aggregate patterns over time, such as to inform the UPLM 150. An example historical user behavior data structure is depicted below by way of illustration:{"user id" : "USER 12345" ,"behavior period" : {" start date" : "2024- 01- 15T00 : 00 : 00Z" ,"end date" : "2024- 11-26T23 : 59 : 59Z"} ,"total content generated" : 1247 ,"total feedback events" : 247 ," feedback rate" : 0 . 198 ," record type distribution" : {" consultation note" : 456 ,"progres s note" : 312 ,"after visit summary" : 289 ,"dis charge summary" : 190} ,"editing patterns" : [{"pattern_id" : " PATTERN_001" ,"pattern type" : " section expansion" ,"target section" : "as ses sment" ," frequency" : 45 ,"average expansion length" : 127 ," common additions" : ["quantitative measurements" ,"di fferential diagnosis" ," clinical reasoning"] ," record types" : [ " consultation note" , "progres s note" ]} ,{"pattern_id" : " PATTERN_002 " ,"pattern type" : "terminology simpli fication" ,"target section" : "patient instructions" ,Docket #: 1040-05WO / US" frequency" : 22 ," common replacements" : [{ " from" : "myocardial infarction" , "to" : "heart attack" } , { " from" : "hypertension" , "to" : "high blood pres sure" } ,{ " from" : "anticoagulation" , "to" : "blood thinner" }] ," record types" : [ "after visit summary" , "dis charge summary" ]} ,{"pattern_id" : " PATTERN_003" ,"pattern type" : "add mis sing field" ,"target fields" : [ "ej ection fraction" , "nyha clas s " , "medication dosages" ] ," frequency" : 38 ," record types" : [ " consultation note" ]}] ,"temporal patterns" : {"time of day preferences" : {"morning" : { "detail level" : 0 . 92 , " formality" : 0 . 75 } , "afternoon" : { "detail level" : 0 . 85 , " formality" : 0 . 72 } , "evening" : { "detail level" : 0 . 78 , " formality" : 0 . 68 }} ,"day of week patterns" : {"monday" : { "average edits per note" : 2 . 3 } ," friday" : { "average edits per note" : 1 . 8 }}} ," response patterns" : {"average time to first edit" : 52 ,"average total editing time" : 165 ,"average edits per content" : 1 . 9 ," confirmation acceptance rate" : 0 . 87} ," satis faction metrics" : {"average satis faction rating" : 4 . 2 ," satis faction s cale" : 5 ," satis faction trend" : "improving" ,"trend slope" : 0 . 15}}

[0118] Prompt selection rules 350 may, for example, be stored in a structured format. The format may, for example, enable dynamic selection based on context. An example prompt selection rule data structure may, for example, include:{" rule_set_id" : "RULESET_USER_12345" ,"user id" : "USER 12345" ,"last updated" : "2024- 11-26T14 : 25 : 00Z" ,"version" : " 3 . 7" ," selection rules" : [{" rule_id" : " SEL_RULE_001" ,"priority" : 1 ," conditions " : {" record type" : " consultation note" ,"user role" : " Physician" ,"department" : "Cardiology" ,Docket #: 1040-05WO / US" content contains" : [ "heart failure" , " cardiomyopathy" ] } ,"action" : {" s elect_prompt " : " PROMPT_CN_003_CARDIAC_ENHANCED " , "apply modi fiers" : ["include quantitative measurements" ,"expand as ses sment section" ,"include ej ection fraction"]} ," confidence" : 0 . 92} ,{" rule_id" : " SEL_RULE_002 " ,"priority" : 2 ," conditions " : {" record type" : "after visit summary" ,"patient age" : ">65" ," content complexity" : "high"} ,"action" : {" select_prompt" : " PROMPT_AVS_012_SIMPLI FIED" , "apply modi fiers" : [" simpli fy terminology" ,"add visual aids" ,"expand follow up instructions"]} ," confidence" : 0 . 85} ,{" rule_id" : " SEL_RULE_003" ,"priority" : 3 ," conditions " : {" record type" : "dis charge summary" ,"length of stay" : ">7 days" ," complications present" : true} ,"action" : {" select_prompt" : " PROMPT_DS_007_DETAILED" ,"apply modi fiers" : ["expand hospital course" ,"detail complications" ," comprehensive medication reconciliation"]} ," confidence" : 0 . 88}] ,"default rules" : {" consultation note" : " PROMPT ON 001" ,"progres s note" : " PROMPT PN 001" ,"after_visit_summary" : " PROMPT_AVS_001" , "dis charge_summary" : " PROMPT_DS_001"} ," rule application statistics" : {"total applications" : 1247 ," succes s ful applications" : 1189 ," succes s rate" : 0 . 953 ,"average confidence" : 0 . 87Docket #: 1040-05WO / US}}

[0119] A PTDP 605 may, for example, aggregate multiple data sources for training. An example PTDP data structure is depicted below by way of illustration:{"ptdp_id" : " PTDP_67890" ,"user id" : "USER 12345" ," created timestamp" : "2024- 11-26T14 : 30 : 00Z" ,"training purpose" : "update prompt for consultation notes" ,"user initiated feedback" : {" feedback ids" : [ " FB 98765" , " FB 98764" , " FB 98763" ] ,"aggregated instructions" : [{"instruction category" : "expand content" ," frequency" : 3 ," common targets" : [ "as ses sment" , "history of present illnes s" ] , " common additions" : [ "quantitative measurements" , " clinical reasoning" ]}] ,"aggregated rules" : [{" rule category" : "include field" ," frequency" : 3 ," common fields" : [ "ej ection fraction" , "nyha clas s" ]}]} ,"user profile" : {"user speci fic weightings" : {"detail level" : 0 . 90 ,"technical terminology" : 0 . 90 ,"include quantitative data" : 0 . 80} ," record type preferences" : {" consultation note" : {"auto expand sections" : [ "as ses sment" , "di fferential diagnosis" ] }}} ,"historical user behavior" : {" relevant patterns" : [{"pattern_id" : " PATTERN_001" ,"pattern type" : " section expansion" ,"target section" : "as ses sment" ," frequency" : 45} ,{"pattern_id" : " PATTERN_003" ,"pattern type" : "add mis sing field" ,"target fields" : [ "ej ection fraction" ] ," frequency" : 38}] ," satis faction trend" : "improving" ,"average satis faction rating" : 4 . 2} ,Docket #: 1040-05WO / US" record type" : " consultation note" ,"trans formed content type" : " clinical note detailed" ,"training features" : {" feature vector" : [ 0 . 90 , 0 . 90 , 0 . 80 , 0 . 75 , 0 . 88 , 0 . 85 ] ," feature names" : ["detail level" ,"technical terminology" ,"include quantitative data" ," formality" ,"include di fferential diagnosis" ,"include treatment rationale"] ," feature weights" : [ 0 . 25 , 0 . 15 , 0 . 20 , 0 . 10 , 0 . 15 , 0 . 15 ]} ," context metadata" : {"department" : "Cardiology" ,"user role" : " Physician" ,"experience level" : " senior" ,"typical patient complexity" : "high"}}

[0120] An updated predefined prompt 615 may be stored, such as with versioning and / or metadata. An example updated predefined prompt data structure is depicted below by way of illustration:{"prompt_id" : " PROMPT_CN_003_CARDIAC_ENHANCED" ,"base prompt id" : " PROMPT CN 003" ,"version" : "2 . 1" ," created timestamp" : "2024- 11-26T14 : 35 : 00Z" ,"user id" : "USER 12345" ," record type" : " consultation note" ,"prompt text" : "Generate a detailed consultation note based on the provided multimedia content . Structure the note with the following sections : Chief Complaint , History of Present Illnes s , Past Medical History, Medications , Allergies , Physical Examination, As ses sment , and Plan . In the As ses sment section, provide a comprehensive analysis including di fferential diagnosis and clinical reasoning . When dis cus sing cardiac conditions , always include relevant quantitative measurements such as ej ection fraction, troponin levels , or BNP values when available in the source content . Use technical medical terminology appropriate for physician-to-physician communication . Make all sections thoroughly detailed with speci fic clinical findings and measurements . " ,"prompt modi fiers" : [{"modi fier id" : "MOD 001" ,"modi fier type" : " section enhancement" ,"target section" : "as ses sment" ,"instruction" : "Expand the as ses sment section with di fferential diagnosis and clinical reasoning . Include quantitative measurements when dis cus sing cardiac conditions . "} ,{"modi fier id" : "MOD 002" ,"modi fier type" : "include field" ," condition" : {"diagnosis contains" : [ "heart failure" , " cardiomyopathy" ] } ,Docket #: 1040-05WO / US"instruction" : "Ensure ej ection fraction and NYHA class are included when discussing heart failure . "}] ,"parameters " : {"max length" : 2000,"temperature" : 0.7 ,"detail level" : "high" ,"formality" : "professional" ,"include differential diagnosis" : true,"include quantitative data" : true} ,"applicable conditions" : {"user roles" : ["Physician" , "Physician Assistant" ] ,"departments" : ["Cardiology" , "Internal Medicine" ] ,"record types" : ["consultation note" , "progress note" ]} ,"performance metrics" : {"usage count" : 156,"average satisfaction rating" : 4.5,"edit frequency" : 0. 12 ,"average edits per use" : 0. 8} ,"parent prompt id" : "PROMPT CN 003" ,"derivation metadata" : {"derived from feedback count" : 3,"learning confidence" : 0. 89,"last training date" : "2024-11-26T14 : 30 : 00Z"}}

[0121] These data structures may, by way of example and not limitation, be stored in various formats including JSON, XML, relational database tables, and / or NoSQL document stores, depending on the implementation. The structures may, for example, be indexed and / or adjusted, such as for efficient retrieval and updating by the UPLM 150 and / or related system components.

[0122] Although various example systems have been described with reference to the FIGS. 1-3, other implementations may be deployed in other industrial, scientific, medical, commercial, and / or residential applications.

[0123] For example, the ACGS 100 and / or the CCGS 200 may be applied to various industries. For example, in legal practice management, the ACGS 100 may be deployed to transform drafts of client interviews into case summaries to be reviewed by the general partner by applying the predefined prompt 140. For example, in finance and accounting, the CCGS 200 may integrate asynchronous transaction notes, audit findings, and analyst commentary into unified reports that are routed among compliance officers and portfolio managers based on user relationships and workflow information. For example, in education or corporate training, the CCGS 200 may be integrated with the ACGS 100 to dynamically generate and integrate personalized lesson summaries that is most effective for each student.Docket #: 1040-05WO / US

[0124] For example, the ACGS 100 may be deployed in legal practice management. For example, the ACGS 100 may be applied to transform client interview recordings into structured case summaries. In this implementation, the multimedia content 105 may include audio recordings of client consultations, deposition transcripts, and / or video recordings of witness interviews. The note classifier 120 may classify the multimedia content 105 into legal record types such as initial client intake notes, case strategy memoranda, discovery summaries, and / or court filing drafts.

[0125] The predefined prompt 140 may be adapted to legal-specific transformations. For example, the predefined prompt 140 may instruct the note editing LLM 125 to “identify potential causes of action,” “extract relevant dates and statute of limitations deadlines,” “flag privileged communications,” and / or “format content according to local court rules.” The user profile 145 may store attorney-specific preferences such as preferred citation formats (e.g., Bluebook vs. ALWD), writing style preferences (e.g., formal vs. conversational tone), and / or j uri sdi cti on- specifi c terminol ogy .

[0126] The UPLM 150 may learn from user-initiated feedback 155. For example, feedback may include when an attorney repeatedly modifies case summaries to emphasize certain legal theories and / or consistently reformats citations. For example, if a partner attorney frequently adds detailed procedural history sections to litigation memoranda, the UPLM 150 may update the user profile 145 to automatically include expanded procedural context in future documents of that record type. The system may generate the transformed content 135 as a polished legal memorandum ready for partner review, even without retraining and / or fine-tuning of the LLM 125.

[0127] The CCGS 200 may enable asynchronous collaboration among legal team members. For example, an intake coordinator may have an introductory phone call and / or video meeting with a potential client. A paralegal may initiate a case file by uploading client intake forms as asynchronous input content (e.g., data input 210). The data integration engine 220 may identify stored records related to the same client or matter and integrate the new intake information with existing case files. The association data store 235 may include legal-specific relationships such as attorney-paralegal assignments, case team structures, court deadlines, and / or matter- specific workflow stages (e.g., pleading stage, discovery stage, motion practice, trial preparation).

[0128] The INDE 230 may, for example, automatically route the integrated content 215 based on legal workflow rules stored in the association data store 235. For example, when a paralegalDocket #: 1040-05WO / UScompletes initial research on a motion, the INDE 230 may automatically deliver the integrated content 215 to the assigned associate attorney for drafting. When the associate completes the draft, the system may route it to the supervising partner, such as based on the internal relationships 710 stored in the user relationships 705. The workflow information 725 may, by way of example and not limitation, include court filing deadlines, hearing dates, and / or discovery cutoff dates, which may, for example, facilitate timely routing and completion of legal work product.

[0129] For example, the ACGS 100 and CCGS 200 may be deployed in finance and / or accounting environments, such as, for example, to transform transaction records, audit findings, and / or financial analysis into unified reports. In this implementation, the multimedia content 105 may include spreadsheet data, scanned receipts, voice memos from client meetings, and / or video recordings of audit interviews. The note classifier 120 may classify the multimedia content 105 into financial record types, such as, for example, transaction logs, audit work papers, compliance reports, tax documentation, and / or portfolio analysis summaries.

[0130] The predefined prompt 140 may, for example, be adapted to financial-specific transformations. For example, the predefined prompt 140 may instruct the LLM 125 to “reconcile transaction amounts with supporting documentation,” “identify potential compliance violations,” “format numbers according to GAAP standards,” and / or “flag transactions exceeding materiality thresholds.” The user profile 145 may store accountantspecific preferences such as preferred financial statement formats, rounding conventions, or industry-specific terminology (e.g., banking vs. manufacturing accounting).

[0131] The UPLM 150 may learn from user-initiated feedback 155, such as when an auditor repeatedly adds detailed variance explanations and / or consistently reformats financial tables. For example, if a senior auditor frequently expands sections discussing internal control weaknesses, the UPLM 150 may update the user profile 145 to automatically include more detailed control testing results in future audit work papers. The system may generate the transformed content 135 as a formatted audit report ready for client review, without demanding retraining of the LLM 125.

[0132] The CCGS 200 may enable asynchronous collaboration among accounting team members. For example, a staff accountant may initiate a quarterly close process, such as by uploading preliminary transaction data as asynchronous input content (e.g., data input 210). The data integration engine 220 may identify stored records from previous quarters andDocket #: 1040-05WO / USintegrate the new transaction data with historical financial records. The association data store 235 may include finance-specific relationships such as accountant-manager-partner hierarchies, client engagement teams, regulatory filing deadlines, and workflow stages (e.g., data collection, reconciliation, review, approval, filing).

[0133] The INDE 230 may automatically route the integrated content 215 based on financial workflow rules stored in the association data store 235. For example, when a staff accountant completes initial transaction coding, the INDE 230 may automatically deliver the integrated content 215 to a senior accountant for reconciliation review. When reconciliation is complete, the system may, for example, route the content to a manager for analytical review, and subsequently to a partner for final approval based on the internal relationships 710. The workflow information 725 may, for example, include SEC filing deadlines, tax return due dates, and / or audit committee meeting schedules, which may, for example, facilitate timely completion and delivery of financial reports.

[0134] The association rules 730 may define conditional routing behaviors specific to financial operations. For example, a rule may specify that “if a transaction variance exceeds 5% of budgeted amount, route immediately to the engagement manager for review.” Another rule may specify that “if audit findings indicate material weakness in internal controls, alert the engagement partner and compliance officer simultaneously.” These rules may, for example, advantageously enable the CCGS 200 to automatically escalate significant financial matters to appropriate personnel without manual intervention.

[0135] For example, the ACGS 100 and CCGS 200 may be deployed in education and / or corporate training environments, such as to generate personalized learning materials and / or track student progress across multiple instructors and / or learning modules. In this implementation, the multimedia content 105 may include lecture recordings, student assignment submissions, quiz responses, and / or video recordings of student presentations. The note classifier 120 may, for example, classify the multimedia content 105 into educational record types such as lesson plans, student progress reports, assessment summaries, and / or individualized education plans (lEPs).

[0136] The predefined prompt 140 may be adapted to education-specific transformations. For example, the predefined prompt 140 may instruct the LLM 125 to “identify learning gaps based on assessment results,” “generate practice problems at appropriate difficulty level,” “summarize key concepts in student-friendly language,” and / or “align content with specific learning objectives or standards.” The user profile 145 may, for example, store instructor-Docket #: 1040-05WO / USspecific preferences such as preferred pedagogical approaches, grading rubrics, or subjectspecific terminology. The predefined prompt 140 may, for example, be student-focused, such as preferred learning styles, preferred instructor types, subject-specific understanding levels, and / or specific summary types.

[0137] The UPLM 150 may learn from user-initiated feedback 155. For example, the UPLM may learn when an instructor repeatedly modifies lesson summaries to include more visual examples and / or consistently adjusts the reading level of generated content. For example, if an instructor frequently simplifies technical vocabulary in computer science lessons for introductory students, the UPLM 150 may update the user profile 145 to automatically generate content with simplified terminology for that instructor’s introductory courses. The system may generate the transformed content 135 as a personalized lesson summary tailored to each student’s learning level, without demanding retraining of the LLM 125.

[0138] The CCGS 200 may enable asynchronous collaboration among educational team members. For example, a teaching assistant may initiate a student progress report by uploading quiz results and / or attendance records as asynchronous input content (e.g., data input 210). The data integration engine 220 may identify stored records from previous assessments and integrate the new performance data with the student’s historical academic record. The association data store 235 may include education-specific relationships such as teacher-student assignments, co-teaching arrangements, parent-teacher communication channels, and / or workflow stages (e.g., instruction, assessment, feedback, remediation, advancement).

[0139] The INDE 230 may automatically route the integrated content 215, such as based on educational workflow rules stored in the association data store 235. For example, when a teaching assistant completes grading of a module assessment, the INDE 230 may automatically deliver the integrated content 215 to the lead instructor for review. If a student’ s performance indicates significant learning gaps, the system may route a customized intervention plan to a learning specialist or tutor based on the internal relationships 710. The workflow information 725 may include assignment due dates, parent-teacher conference schedules, and / or standardized testing windows, which may advantageously facilitate timely communication and / or intervention.

[0140] The association rules 730 may define conditional routing behaviors specific to educational operations. For example, a rule may specify that “if a student scores below 70% on two consecutive assessments, automatically generate an intervention plan and route to the student’s advisor and parents.” Another rule may specify that “if a student demonstratesDocket #: 1040-05WO / USmastery of all learning objectives in a unit, route advanced enrichment materials to the student’s learning portal.” These rules may, for example, advantageously enable the CCGS 200 to provide personalized, adaptive learning experiences. For example, the learning experiences may respond to individual student needs without relying on manual tracking by instructors.

[0141] The system may integrate data from multiple sources across a student’s educational journey. For example, the data integration engine 220 may combine formative assessment data from daily classroom activities, summative assessment results from unit tests, behavioral observations from multiple teachers, and / or engagement metrics from online learning platforms. The INDE 230 may then route comprehensive progress reports to appropriate stakeholders (e.g., parents, counselors, administrators) based on the external relationships 715 stored in the association data store 235.

[0142] For example, the ACGS 100 and CCGS 200 may be deployed in manufacturing and quality control environments to transform inspection reports, equipment maintenance logs, and / or production data into actionable quality assurance documentation. In this implementation, the multimedia content 105 may include images of product defects, audio recordings of equipment malfunction descriptions, sensor data from production lines, and / or video recordings of manufacturing processes. The note classifier 120 may classify the multimedia content 105 into manufacturing record types such as, by way of example and not limitation, quality inspection reports, corrective action requests, equipment maintenance records, and / or production batch documentation.

[0143] The predefined prompt 140 may be adapted to manufacturing-specific transformations. For example, the predefined prompt 140 may instruct the note editing LLM 125 to “identify root causes of quality defects,” “correlate defect patterns with production parameters,” “generate corrective action recommendations based on ISO 9001 standards,” and / or “format content according to FDA regulatory guidelines.” The user profile 145 may store, for example, quality engineer-specific preferences such as, by way of example and not limitation, preferred defect classification systems, statistical analysis methods, and / or industry-specific terminology (e.g., Six Sigma, Lean Manufacturing).

[0144] The UPLM 150 may learn from user-initiated feedback 155. For example, feedback may include when a quality manager repeatedly adds detailed process capability analysis and / or consistently expands sections discussing preventive measures. For example, if a quality director frequently adds supplier corrective action items to defect reports, the UPLM 150 mayDocket #: 1040-05WO / USupdate the user profile 145 to automatically include supplier notification sections in future quality reports. The system may generate the transformed content 135 as a comprehensive quality report ready for regulatory submission, without retraining of the LLM 125.

[0145] The CCGS 200 may enable asynchronous collaboration among manufacturing team members. For example, a production line inspector may initiate a quality incident report by uploading defect images and measurements as asynchronous input content (e.g., data input 210). The data integration engine 220 may identify stored records from previous quality incidents involving similar defect types and integrate the new incident data with historical quality metrics. The association data store 235 may include manufacturing-specific relationships such as inspector-supervisor-quality manager hierarchies, cross-functional corrective action teams, supplier relationships, and workflow stages (e.g., detection, containment, root cause analysis, corrective action, verification).

[0146] The INDE 230 may automatically route the integrated content 215 based on manufacturing workflow rules stored in the association data store 235. For example, when an inspector identifies an urgent defect, the INDE 230 may immediately deliver the integrated content 215 to the production supervisor and quality manager simultaneously. If root cause analysis implicates a supplier component, the system may automatically route a supplier corrective action request to the procurement team based on the external relationships 715. The workflow information 725 may include production schedules, regulatory audit dates, and / or customer delivery commitments, such as to facilitate timely resolution of quality issues.

[0147] The association rules 730 may define conditional routing behaviors specific to manufacturing operations. For example, a rule may specify that “if defect rate exceeds 2% for any product line, immediately halt production and route emergency notification to plant manager and quality director.” Another rule may specify that “if the same defect type recurs within 30 days, escalate to executive leadership and initiate formal corrective action process.” Such rules may, for example, advantageously enable the CCGS 200 to provide rapid response to quality issues and / or facilitate compliance with regulatory frameworks without manual monitoring.

[0148] Various embodiments may therefore support adaptive and asynchronous collaboration across a wide range of industrial, scientific, medical, commercial, and residential contexts, enabling organizations to improve efficiency, reduce errors, and maintain compliance while adapting to user preferences without the computational burden of retraining large language models.Docket #: 1040-05WO / US

[0149] Various embodiments provide specific technical improvements over conventional content transformation systems that address concrete technical problems in computer functionality and machine learning deployment.

[0150] For example, conventional systems may rely on full or partial retraining or adaptation (e.g., fine-tuning) of large language models to adapt to user preferences. Model retraining and / or fine-tuning is computationally expensive in terms of processing resources, energy consumption, and time. For example, retraining a large language model may consume substantial GPU hours, consume significant electrical energy, and / or may take extended periods (e.g., days, weeks, or months) to complete. This computational burden makes realtime and / or near-real-time personalization practically infeasible in conventional systems.

[0151] Various embodiments (e.g., of the ACGS 100) may advantageously solve this technical problem. For example, embodiments may implement a user preference learning module (UPLM 150). The UPLM may, for example, operate on a limited scope of user-specific data rather than retraining the note editing LLM 125. As described with reference to FIG. 1, the UPLM 150 may, for example, modify (e.g., only) the predefined prompt 140 and / or prompt selection rules 350 (e.g., by modifying the prompt itself and / or through ‘add-on’ instructions that may dynamically - such as at content transformation time - modify the prompt submitted to the LLM). For example, the ACGS may personalize the LLM outputs without modifying model weights (e.g., of the LLM 125). This architectural separation may, for example, advantageously enable user-level adaptation while reducing computational overhead.

[0152] The system may, for example, achieve this efficiency by modifying input parameters to the LLM 125 (e.g., via modification and / or replacement of the predefined prompt 140) without modifying the model weights. The predefined prompt 140 may, for example, direct the output of the pre-trained LLM 125 through dynamically adjusted prompts that encode user preferences stored in the user profile 145. The user profile 145 may, for example, include userspecific weightings for selecting the predefined prompt 140. These weights may, for example, demand significantly less storage and / or processing than generating, maintaining, and / or applying separate fine-tuned models for each user.

[0153] The UPLM 150 may, for example, generate prompt-adjustment parameters based on user-initiated feedback 155 received from the user device 130. These prompt-adjustment parameters may, for example, update the user profile 145 and / or the predefined prompt 140 for later use (e.g., without retraining the large language model 125). This approach may, forDocket #: 1040-05WO / USexample, advantageously provide personalized content transformation while avoiding the computational costs associated with model retraining and / or fine-tuning.

[0154] Electronic medical record systems may, for example, rely on manual routing of records between users. Such systems may, for example, often fail to maintain context across asynchronous contributions from multiple devices. This may, for example, create technical problems including increased network traffic from redundant data retrieval, processing delays from sequential workflow execution, and / or data inconsistency from lack of synchronized updates.

[0155] Various embodiments may, for example, advantageously provide technical solutions such as via association-based automatic routing, such as with stored context. For example, in some embodiments (e.g., implementing one or more CCGS 200) may, for example, solve technical problems through an integrated notes delivery engine(s) (e.g., INDE 230) and / or association data store 235, such as described with reference to FIG. 2. The association data store 235 may, for example, maintain relationships between users, record types, and / or workflow stages. These data relationships may, for example, advantageously enable the INDE 230 to automatically determine target destinations, even without manual user input.

[0156] As described with reference to FIG. 7, the association data store 235 may, for example, include user relationships 705 (e.g., including internal relationships 710 and / or external relationships 715), record types 720, workflow information 725, and / or association rules 730. The INDE 230 may, for example, use these stored associations to automatically route the integrated content 215 to the appropriate target destination based on the determined association and current situation.

[0157] For example, when the data integration engine 220 receives asynchronous input content (e.g., data input 210) from a first user device, it may identify stored records associated with the asynchronous input content using the association data store 235. The system may, for example, generate transformed content based on both the asynchronous input content and the stored record. The INDE 230 may, for example, automatically route the integrated content 215 to the next user in the workflow, such as based on the associations stored in the association data store 235. This may, for example, advantageously reduce or eliminate manual routing steps and / or enable asynchronous collaboration 225 across multiple users 205 and / or devices.

[0158] The workflow information 725 may, for example, include temporal sequences defining order and / or timing of actions, scheduling information, patient status indicators, and / or pending action flags. These elements may, for example, advantageously enable the system toDocket #: 1040-05WO / USmaintain continuity and / or traceability throughout asynchronous workflows, such as without synchronous user interaction.

[0159] Conventional personalized Al systems may, for example, face a scalability problem: as the number of users increases, the computational cost and storage demands of maintaining per-user models grow (e.g., linearly, super-linearly). Systems that maintain individually finetuned models for each user may, for example, consume substantial storage and / or computational resources that scale with the number of users.

[0160] Various embodiments may, for example, solve this technical problem with shared models with user-specific profiles. For example, ACGS embodiments (e.g., ACGS 100) may advantageously provide a technical solution to solve computational scalability problems by maintaining a single shared LLM 125 while storing lightweight user profiles 145 for each user. As described with reference to FIG. 3, for example, the user profile 145 may be stored in the data store 345. The user profile 145 may, for example, include user-specific weightings, such as for selecting a predefined prompt 140. This approach may, for example, demand significantly less storage per user compared to maintaining separate fine-tuned models.

[0161] When the NGE 115 receives multimedia content 105, it may, for example, retrieve the relevant user profile 145. It may use, for example, the relevant user profile to select the predefined prompt 140, such as based on the prompt selection rules 350. The predefined prompt 140 may, for example, be applied to the shared LLM 125 to generate the adaptively transformed content 135. This architecture may, for example, advantageously enable the system to scale to large numbers of users while maintaining a single shared model infrastructure, such as advantageously reducing per-user storage and / or computational demands.

[0162] Various embodiments may, for example, provide improvements in computer functionality. For example, some embodiments may advantageously reduce computational overhead. Prompt adaptation (e.g., through the UPLM 150) may, for example, advantageously avoid the computational burden of model retraining and / or fine-tuning. This may, for example, advantageously enable user-level personalization without the processing costs associated with retraining and / or fine-tuning the LLM 125.

[0163] Various embodiments may, for example, improve energy efficiency. For example, by avoiding model retraining, the system may reduce energy consumption associated with adaptation cycles.Docket #: 1040-05WO / US

[0164] Various embodiments may, for example, reduce latency. For example, the association data store 235 and automatic routing through the INDE 230 may advantageously reduce processing delays, such as by eliminating manual routing steps and / or enabling parallel asynchronous processing.

[0165] Various embodiments may, for example, enhance scalability. For example, per-user storage demands may be advantageously reduced by storing lightweight user profiles 145 rather than per-user fine-tuned models. Such embodiments may, for example, advantageously enable computationally practical scaling to large numbers of users.

[0166] Various embodiments may, for example, improve computational workflow efficiency. Asynchronous integration, such as through the data integration engine 220 with stored association data, may, for example, advantageously reduce redundant operations and / or enable incremental building of electronic records across multiple contributors.

[0167] In various embodiments, some bypass circuits implementations may be controlled in response to signals from analog or digital components, which may be discrete, integrated, or a combination of each. Some embodiments may include programmed, programmable devices, or some combination thereof (e.g., PLAs, PLDs, ASICs, microcontroller, microprocessor), and may include one or more data stores (e.g., cell, register, block, page) that provide single or multi-level digital data storage capability, and which may be volatile, non-volatile, or some combination thereof. Some control functions may be implemented in hardware, software, firmware, or a combination of any of them.

[0168] Computer program products may contain a set of instructions that, when executed by a processor device, cause the processor to perform prescribed functions. These functions may be performed in conjunction with controlled devices in operable communication with the processor. Computer program products, which may include software, may be stored in a data store tangibly embedded on a storage medium, such as an electronic, magnetic, or rotating storage device, and may be fixed or removable (e.g., hard disk, floppy disk, thumb drive, CD, DVD).

[0169] Although an example of a system, which may be portable, has been described with reference to the above figures, other implementations may be deployed in other processing applications, such as desktop and networked environments.

[0170] Temporary auxiliary energy inputs may be received, for example, from chargeable or single use batteries, which may enable use in portable or remote applications. Some embodiments may operate with other DC voltage sources, such as (nominal) batteries, forDocket #: 1040-05WO / USexample. Alternating current (AC) inputs, which may be provided, for example from a 50 / 60 Hz power port, or from a portable electric generator, may be received via a rectifier and appropriate scaling. Provision for AC (e.g., sine wave, square wave, triangular wave) inputs may include a line frequency transformer to provide voltage step-up, voltage step-down, and / or isolation.

[0171] Although particular features of an architecture have been described, other features may be incorporated to improve performance. For example, caching (e.g., LI, L2, ...) techniques may be used. Random access memory may be included, for example, to provide scratch pad memory and or to load executable code or parameter information stored for use during runtime operations. Other hardware and software may be provided to perform operations, such as network or other communications using one or more protocols, wireless (e.g., infrared) communications, stored operational energy and power supplies (e.g., batteries), switching and / or linear power supply circuits, software maintenance (e.g., self-test, upgrades), and the like. One or more communication interfaces may be provided in support of data storage and related operations.

[0172] Some systems may be implemented as a computer system that can be used with various implementations. For example, various implementations may include digital circuitry, analog circuitry, computer hardware, firmware, software, or combinations thereof. Apparatus can be implemented in a computer program product tangibly embodied in an information carrier, e.g., in a machine-readable storage device, for execution by a programmable processor; and methods can be performed by a programmable processor executing a program of instructions to perform functions of various embodiments by operating on input data and generating an output. Various embodiments can be implemented advantageously in one or more computer programs that are executable on a programmable system including at least one programmable processor coupled to receive data and instructions from, and to transmit data and instructions to, a data storage system, at least one input device, and / or at least one output device. A computer program is a set of instructions that can be used, directly or indirectly, in a computer to perform a certain activity or bring about a certain result. A computer program can be written in any form of programming language, including compiled or interpreted languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0173] Suitable processors for the execution of a program of instructions include, by way of example, both general and special purpose microprocessors, which may include a singleDocket #: 1040-05WO / USprocessor or one of multiple processors of any kind of computer. Generally, a processor will receive instructions and data from a read-only memory or a random-access memory or both. The essential elements of a computer are a processor for executing instructions and one or more memories for storing instructions and data. Generally, a computer will also include, or be operatively coupled to communicate with, one or more mass storage devices for storing data files; such devices include magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and optical disks. Storage devices suitable for tangibly embodying computer program instructions and data include all forms of non-volatile memory, including, by way of example, semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, ASICs (application-specific integrated circuits).

[0174] In some implementations, each system may be programmed with the same or similar information and / or initialized with substantially identical information stored in volatile and / or non-volatile memory. For example, one data interface may be configured to perform auto configuration, auto download, and / or auto update functions when coupled to an appropriate host device, such as a desktop computer or a server.

[0175] In some implementations, one or more user-interface features may be custom configured to perform specific functions. Various embodiments may be implemented in a computer system that includes a graphical user interface and / or an Internet browser. To provide for interaction with a user, some implementations may be implemented on a computer having a display device. The display device may, for example, include an LED (light-emitting diode) display. In some implementations, a display device may, for example, include a CRT (cathode ray tube). In some implementations, a display device may include, for example, an LCD (liquid crystal display). A display device (e.g., monitor) may, for example, be used for displaying information to the user. Some implementations may, for example, include a keyboard and / or pointing device (e.g., mouse, trackpad, trackball joystick), such as by which the user can provide input to the computer.

[0176] In various implementations, the system may communicate using suitable communication methods, equipment, and techniques. For example, the system may communicate with compatible devices (e.g., devices capable of transferring data to and / or from the system) using point-to-point communication in which a message is transported directly from the source to the receiver over a dedicated physical link (e.g., fiber optic link,Docket #: 1040-05WO / USpoint-to-point wiring, daisy-chain). The components of the system may exchange information by any form or medium of analog or digital data communication, including packet-based messages on a communication network. Examples of communication networks include, e.g., a LAN (local area network), a WAN (wide area network), MAN (metropolitan area network), wireless and / or optical networks, the computers and networks forming the Internet, or some combination thereof. Other implementations may transport messages by broadcasting to all or substantially all devices that are coupled together by a communication network, for example, by using omni-directional radio frequency (RF) signals. Still other implementations may transport messages characterized by high directivity, such as RF signals transmitted using directional (e.g., narrow beam) antennas or infrared signals that may optionally be used with focusing optics. Still other implementations are possible using appropriate interfaces and protocols such as, by way of example and not intended to be limiting, USB 2.0, Firewire, ATA / IDE, RS-232, RS-422, RS-485, 802.11 a / b / g, Wi-Fi, Ethernet, IrDA, FDDI (fiber distributed data interface), token-ring networks, multiplexing techniques based on frequency, time, or code division, or some combination thereof. Some implementations may optionally incorporate features such as error checking and correction (ECC) for data integrity, or security measures, such as encryption (e.g., WEP) and password protection.

[0177] In various embodiments, the computer system may include Internet of Things (loT) devices. loT devices may include objects embedded with electronics, software, sensors, actuators, and network connectivity which enable these objects to collect and exchange data. loT devices may be in-use with wired or wireless devices by sending data through an interface to another device. loT devices may collect useful data and then autonomously flow the data between other devices.

[0178] Various examples of modules may be implemented using circuitry, including various electronic hardware. By way of example and not limitation, the hardware may include transistors, resistors, capacitors, switches, integrated circuits, other modules, or some combination thereof. In various examples, the modules may include analog logic, digital logic, discrete components, traces and / or memory circuits fabricated on a silicon substrate including various integrated circuits (e.g., FPGAs, ASICs), or some combination thereof. In some embodiments, the module(s) may involve execution of preprogrammed instructions, software executed by a processor, or some combination thereof. For example, various modules may involve both hardware and software.Docket #: 1040-05WO / US

[0179] Some embodiments may, for example, desensitize and / or resensitize data as disclosed at least with reference to WO patent application serial no. PCT / US2024 / 035272, filed Jun 24, 2024 (naming inventor(s) including SINGH, Simerjot) and titled "Dynamically generated LLM request package", the entire contents of which are incorporated herein by reference.

[0180] Some embodiments may, for example, generate and / or interact with media (e.g., multimedia) such as disclosed at least with reference to WO patent application serial no. US2020033328, filed May 17, 2020 (naming inventor(s) including COURT, Kenneth) and titled "Apparatus for generating and transmitting annotated video sequences in response to manual and image input devices", the entire contents of which are incorporated herein by reference.

[0181] Some embodiments may, for example, provide access to secure data such as disclosed at least with reference to WO patent application serial no. PCT / US2022 / 072888, filed Jun 10, 2022 (naming inventor(s) including ODLAND, Gregory) and titled "Multi-party controlled transient user credentialing for interaction with patient health data", the entire contents of which are incorporated herein by reference.

[0182] In a first illustrative aspect, a computer-implemented method may be used to automatically generate output content through a dynamically adjusted transformation process.

[0183] For example, the method may include receiving a signal to generate an adaptively transformed content.

[0184] For example, the method may include classifying a multimedia content into one of several record types.

[0185] For example, the method may include selecting a predefined prompt based on at least the record type and a user profile.

[0186] For example, the user profile may be associated with a user of the adaptively transformed content.

[0187] For example, the method may include applying a large language model with the predefined prompt to generate the adaptively transformed content.

[0188] For example, the method may include applying at least one user-initiated feedback received from a user device to a user preference learning model to generate prompt-adjustment parameters.

[0189] For example, the method may include updating the user profile as a function of the prompt-adjustment parameters.Docket #: 1040-05WO / US

[0190] For example, the update may modify at least one predefined prompt or promptselection rule to produce a refined prompt.

[0191] For example, the refined prompt may be used for subsequent multimedia content related to the user profile and the record type without retraining the large language model.

[0192] In some examples, the adaptive electronic records may include electronic medical records.

[0193] In some examples, the user-initiated feedback may include editing-operation instructions.

[0194] In some examples, the at least one user-initiated feedback may include at least one prompt inputted by the user in response to the adaptively transformed content generated in response to the predefined prompt.

[0195] In some examples, the user-initiated feedback may include evaluation instructions.

[0196] In some examples, the user-initiated feedback may include user-defined rules.

[0197] In some examples, the user preference learning model may operate on a limited scope of input that includes user-generated instructions.

[0198] In some examples, the predefined prompt may include artificial intelligence-generated suggestions for editing the multimedia content.

[0199] In some examples, the user profile may include user-specific weightings for selecting the predefined prompt.

[0200] Applying the at least one user-initiated feedback to the user preference learning model may include receiving editing operations from the user device. Each editing operation may include an operation type, a location within the adaptively transformed content, and modified text. Applying the at least one user-initiated feedback may include generating the promptadjustment parameters based on the editing operations. The prompt-adjustment parameters may include weight adjustments including a previous value, an adjustment delta, a new value, and a confidence score.

[0201] The user preference learning model may operate on a limited scope of user-specific data without modifying model weights of the large language model.

[0202] The operations may include aggregating historical user behavior data including editing patterns. Each editing pattern may include a pattern type, a target section, a frequency count, and common additions. The operations may include selecting the predefined prompt based on the historical user behavior data in addition to the at least the record type and the user profile.Docket #: 1040-05WO / US

[0203] The historical user behavior data further may include temporal patterns including time-of-day preferences and day-of-week patterns. Each temporal pattern may associate a time period with user-specific weightings. Selecting the predefined prompt may be further based on a current time corresponding to one of the temporal patterns.

[0204] Selecting the predefined prompt may include applying prompt selection rules including a plurality of selection rules. Each selection rule may include a priority, conditions based on the record type and user role, and an action specifying a prompt identifier and modifiers to apply. Selecting the predefined prompt may include selecting a highest-priority selection rule having conditions satisfied by the multimedia content and the user profile.

[0205] In a second illustrative aspect, an asynchronous content generation system may include a data store that contains a program of instructions and a processor that executes those instructions.

[0206] For example, the processor may perform operations to asynchronously update an electronic content. For example, the processor may receive asynchronous input content that is associated with an existing record type. For example, the processor may identify a stored record associated with the asynchronous input content. For example, the processor may generate a transformed content as a function of the asynchronous input content and the stored record.

[0207] For example, the processor may integrate the transformed content with the stored record to produce an updated integrated content. For example, the processor may determine, based on an association identified within the asynchronous input content, a target destination for the updated integrated content. For example, the processor may transmit the updated integrated content to the target destination.

[0208] In some examples, generating the transformed content may include performing one or more of the computer-implemented method described with reference to the first illustrative aspect.

[0209] In some examples, receiving asynchronous input content may include asynchronously receiving multiple streams of multimedia content from several devices.

[0210] In some examples, determining the target destination may include identifying, from an association data store, a subsequent stage related to the stored record.

[0211] For example, the subsequent stage may also be associated with the target destination.

[0212] In some examples, the subsequent stage may be associated with a user, and the target destination may be a user device linked to that user.Docket #: 1040-05WO / US

[0213] In some examples, the association data store may include associations of user relationships, record types, and workflow information.

[0214] In some examples, integrating the transformed content may include merging multiple transformed contents into the updated integrated content based on the association in the association data store.

[0215] In some examples, the operations further include maintaining an association data store including user relationships, record types, and workflow information. The workflow information may include temporal sequences defining order and timing of actions. Determining the target destination may include identifying, from the association data store, a subsequent workflow stage associated with the stored record and a user assigned to the subsequent workflow stage.

[0216] The association data store may include association rules defining conditional routing behaviors. Each association rule may include a condition based on content of the asynchronous input content and an action specifying a target destination. Determining the target destination may include applying the association rules to the asynchronous input content.

[0217] The operations may include receiving a plurality of asynchronous input contents from a plurality of user devices at different times. The operations may include identifying a common stored record associated with the plurality of asynchronous input contents. The operations may include integrating each of the plurality of asynchronous input contents with the common stored record such that the updated integrated content is incrementally built without relying on synchronous interaction among the plurality of user devices.

[0218] Generating the transformed content may include retrieving a user profile associated with a user of the target destination. The user profile may include user-specific weightings represented as numerical values. Generating the transformed content may include selecting a predefined prompt based on the record type and the user-specific weightings. Generation the transformed content may include applying a large language model with the predefined prompt to generate the transformed content.

[0219] In a third illustrative aspect, a computer program product may include one or more programs of instructions tangibly embodied on non-transitory computer readable medium wherein, when the instructions are executed on a processor, cause the processor to carry out the steps of any of the examples of the first and / or second illustrative aspects.Docket #: 1040-05WO / US

[0220] In a fourth illustrative aspect, a system may include components (e.g., as disclosed herein at least with reference to computing systems) configured to carry out the steps of any of the examples of the first and / or second illustrative aspects.

[0221] A number of implementations have been described. Nevertheless, it will be understood that various modifications may be made. For example, advantageous results may be achieved if the steps of the disclosed techniques were performed in a different sequence, or if components of the disclosed systems were combined in a different manner, or if the components were supplemented with other components. Accordingly, other implementations are contemplated within the scope of the following claims.

Claims

Docket #: 1040-05WO / USCLAIMSWhat is claimed is:

1. A computer-implemented method performed by at least one processor to automatically generate output content based on a dynamically adjusted transformation process, the method comprising:receiving a signal to generate an adaptively transformed content (135); classifying a multimedia content (105) into one of a plurality of record types; selecting a predefined prompt (140) based on at least the record type and a user profile (145), wherein the user profile is associated with a user of the adaptively transformed content;applying a large language model (125) with the predefined prompt to generate the adaptively transformed content;applying at least one user-initiated feedback (155) received from a user device (130) to a user preference learning model (150) to generate prompt-adjustment parameters; and updating the user profile as a function of the prompt-adjustment parameters, such that at least one predefined prompt and / or prompt-selection rules is modified to produce a refined prompt, wherein the refined prompt is applied to subsequent multimedia content associated with the user profile and the at least the record type without retraining the large language model.

2. The computer-implemented method of claim 1, wherein the adaptively transformed content comprises electronic medical records.

3. The computer-implemented method of any of claims 1-2, wherein the at least one user- initiated feedback comprises editing-operation instructions.

4. The computer-implemented method of any of claims 1-3, wherein the at least one user- initiated feedback comprises at least one prompt inputted by the user in response to the adaptively transformed content generated in response to the predefined prompt.Docket #: 1040-05WO / US5. The computer-implemented method of any of claims 1-4, wherein the at least one user- initiated feedback comprises evaluation instructions.

6. The computer-implemented method of any of claims 1-5, wherein the at least one user- initiated feedback comprises user-defined rules.

7. The computer-implemented method of any of claims 1-6, wherein the user preference learning model operates on a limited scope of input comprising user-generated instructions.

8. The computer-implemented method of any of claims 1-7, wherein the predefined prompt comprises artificial intelligence (Al) generated suggestions in editing the multimedia content.

9. The computer-implemented method of any of claims 1-8, wherein the user profile comprises user specific weightings for selecting the predefined prompt.

10. The computer-implemented method of any one of claims 1-9, wherein the user profile comprises a structured data object containing user-specific weightings represented as numerical values, wherein the user-specific weightings quantify user preferences for content characteristics including detail level, formality, and technical terminology preference.

11. The computer-implemented method of any one of claims 1-10, wherein applying the at least one user-initiated feedback to the user preference learning model comprises:receiving editing operations from the user device, wherein each editing operation comprises an operation type, a location within the adaptively transformed content, and modified text; andgenerating the prompt-adjustment parameters based on the editing operations, wherein the prompt-adjustment parameters comprise weight adjustments including a previous value, an adjustment delta, a new value, and a confidence score.

12. The computer-implemented method of any one of claims 1-11, wherein the user preference learning model operates on a limited scope of user-specific data without modifying model weights of the large language model.Docket #: 1040-05WO / US13. The computer-implemented method of any one of claims 1-12, further comprising:aggregating historical user behavior data comprising editing patterns, wherein each editing pattern comprises a pattern type, a target section, a frequency count, and common additions; andselecting the predefined prompt based on the historical user behavior data in addition to the at least the record type and the user profile.

14. The computer-implemented method of claim 13, wherein the historical user behavior data further comprises temporal patterns including time-of-day preferences and day-of-week patterns, wherein each temporal pattern associates a time period with user-specific weightings, and wherein selecting the predefined prompt is further based on a current time corresponding to one of the temporal patterns.

15. The computer-implemented method of any one of claims 1-14, wherein selecting the predefined prompt comprises:applying prompt selection rules comprising a plurality of selection rules, wherein each selection rule comprises a priority, conditions based on the at least the record type and user role, and an action specifying a prompt identifier and modifiers to apply; andselecting a highe st-priori ty selection rule having conditions satisfied by the multimedia content and the user profile.Docket #: 1040-05WO / US16. An asynchronous content generation system (200), comprising:a data store (345) comprising a program of instructions; anda processor (305) operably coupled to the data store such that, when executing the program of instructions, the processor causes operations to be performed to asynchronously update an electronic content, the operations comprising:receiving asynchronous input content (210) associated with a record type; identifying a stored record associated with the asynchronous input content; generating a transformed content as a function of the asynchronous input content and the stored record;integrating the transformed content with the stored record to generate an updated integrated content (215);determining, based on an association identified within the asynchronous input content, a target destination for the updated integrated content; andtransmitting the updated integrated content to the target destination.

17. The asynchronous content generation system of claim 16, wherein generating the transformed content comprises performing the computer-implemented method of any of claims 1-8.

18. The asynchronous content generation system of any one of claims 16-17, wherein receiving asynchronous input content comprises asynchronously receiving multiple streams of multimedia content from a plurality of devices.

19. The asynchronous content generation system of any one of claims 16-18, wherein determining the target destination comprises identifying, from an association data store, a subsequent stage associated with the stored record, wherein the subsequent stage is further associated with the target destination.Docket #: 1040-05WO / US20. The asynchronous content generation system of claim 19, wherein the subsequent stage is associated with the user, wherein the target destination is the user device associated with the user.

21. The asynchronous content generation system of any one of claims 16-20, wherein the association data store comprises associations of user relationships, record types, and workflow.

22. The asynchronous content generation system of claim 21, wherein integrating the transformed content comprises integrating multiple transformed contents into the updated integrated content based on the association in the association data store.

23. The asynchronous content generation system of any one of claims 16-22, wherein the operations further comprise:maintaining an association data store comprising user relationships, record types, and workflow information, wherein the workflow information comprises temporal sequences defining order and timing of actions; andwherein determining the target destination comprises identifying, from the association data store, a subsequent workflow stage associated with the stored record and a subsequent user assigned to the subsequent workflow stage.

24. The asynchronous content generation system of claim 23, wherein the association data store further comprises association rules defining conditional routing behaviors, wherein each association rule comprises a condition based on content of the asynchronous input content and an action specifying the target destination, and wherein determining the target destination comprises applying the association rules to the asynchronous input content.

25. The asynchronous content generation system of any one of claims 16-24, wherein the operations further comprise:receiving a plurality of asynchronous input contents from a plurality of user devices at different times;identifying a common stored record associated with the plurality of asynchronous input contents; andDocket #: 1040-05WO / USintegrating each of the plurality of asynchronous input contents with the common stored record such that the updated integrated content is incrementally built without relying on synchronous interaction among the plurality of user devices.

26. The asynchronous content generation system of any one of claims 16-25, wherein generating the transformed content comprises:retrieving a user profile associated with a user of the target destination, wherein the user profile comprises user-specific weightings represented as numerical values;selecting a predefined prompt based on the record type and the user-specific weightings; andapplying a large language model with the predefined prompt to generate the transformed content.

27. A computer program product comprising a program of instructions tangibly embodied on a non-transitory computer readable medium wherein, when the instructions are executed on a processor, cause the processor to carry out steps of the method of any one of claims 1-26.

28. A system comprising means for carrying out steps of the method of any one of claims 1- 26.

Images