Messaging system decision reapplication engine

Abstract

Provided is a method including receiving a message that is destined to one or more recipients and that includes a plurality of data fields, where a first set of data fields of the plurality of data fields includes unstructured data. The method includes parsing the unstructured data in at least a portion of the first set of data fields to obtain parsed data and processing, with a large language model, the parsed data to determine an interpretation of the data and a confidence score in that interpretation. The method determines, based on the interpretation and the confidence score, one or more actions to take with at least one of the message or the unstructured data in the at least the portion of the first set of data fields, performs the one or more actions, and sends the message to the one or more recipients.

Description

Attorney Docket No. 201818-0585628MESSAGING SYSTEM DECISION REAPPLICATION ENGINEBACKGROUND1. Field

[0001] The present disclosure relates generally to electronic messaging and artificial intelligence, and more specifically to perform decision reapplication of electronic messages and data included in the electronic messages during transmission.2. Description of the Related Art

[0002] Various industries use messaging systems between users to complete various processes. Standards in messages are often used to allow for a common understanding of the data across linguistic and systems boundaries and to permit the seamless, automated transmission, receipt and processing of communications exchanged between users. Use of standardized messages and reference data enables data exchanged between institutions to be unambiguous and machine friendly; in turn this enables efficient automation, thereby reducing costs and mitigating risks. Some standards allow for free text in certain fields to provide instructions or provide other important information that gives the standardized messages flexibility which is vital for the message.SUMMARY

[0003] According to one innovative aspect of the subject matter described in this application, a process includes a non-transitory, machine-readable medium storing instructions that, when executed by one or more processors, effectuate operations that include receiving, by a computing system, a message that is destined to one or more recipients and that includes a plurality of data fields, wherein a first set of data fields of the plurality of data fields includes unstructured data; parsing, by the computing system, the unstructured data in at least a portion of the first set of data fields to obtain parsed data; processing, with a large language model, the parsed data to determine an interpretation of the data and a confidence score in that interpretation; determining, by the computing system and on the interpretation and the confidence score, one or more actions to take with at least one of the message or the unstructured data in the at least the portion of the first setAttorney Docket No. 201818-0585628 of data fields; performing, by the computing system, the one or more actions; and sending, by the computing system, the message to the one or more recipients.

[0004] According to another innovation aspect of the subject matter described in this application a process includes: obtaining, by a computer system, a plurality of historical messages that were sent via a messaging system and that each include a plurality of data fields, wherein a first set of data fields of the plurality of data fields includes unstructured data; obtaining, by the computer system, one or more historical actions taken for a respective historical message of the plurality of historical messages; training, by the computer system and using the one or more historical actions and at least a portion of the unstructured data in at least a portion of the first set of data fields of the plurality data of fields, a large language model; and saving, by the computer system, the large language model in storage.

[0005] According to another innovative aspect of the subject matter described in this application, a non-transitory, machine-readable medium storing instructions that, when executed by one or more processors, effectuate operations that include either of the processes above.

[0006] According to another innovative aspect of the subject matter described in this application a system that includes one or more processors; and memory that stores instructions that when executed by the one or more processors causes the one or more processors to effectuate operations including cither of the processes above.

[0007] The details of one or more implementations of the subject matter described in this specification are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the subject matter will become apparent from the description, the drawings, and the claims.BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The above-mentioned aspects and other aspects of the present techniques will be better understood when the present application is read in view of the following drawings in which like numbers indicate similar or identical elements:Attorney Docket No. 201818-0585628

[0009] FIG. 1 is a block diagram illustrating an example of a messaging system, in accordance with some embodiments of the present disclosure;

[0010] FIG. 2 is a block diagram illustrating an example of a user computing device of the messaging system of FIG. 1, in accordance with some embodiments of the present disclosure;

[0011] FIG. 3 is a block diagram illustrating an example of a message processing computing device of the messaging system of FIG. 1, in accordance with some embodiments of the present disclosure;

[0012] FIG. 4 is a block diagram illustrating a messaging gateway of the messaging system of FIG. 1, in accordance with some embodiments of the present disclosure;

[0013] FIG. 5 is a flow diagram illustrating an example of a method of message decision reapplication performed by the messaging system of FIG. 1, in accordance with some embodiments of the present disclosure;

[0014] FIG. 6 is a flow diagram illustrating a method of training a message decision reapplication model used in the message decision reapplication of FIG. 5, in accordance with some embodiments of the present disclosure;

[0015] FIG. 7 is a workflow of message decision reapplication during the method of FIG. 5 and training the message decision reapplication model during the method of FIG. 6, in accordance with some embodiments of the present disclosure;

[0016] FIG. 8 is a workflow of training an artificial intelligence decision reapplication model and using that model to make message decision during the methods of FIG. 5 and FIG. 6, in accordance with some embodiments of the present disclosure;

[0017] FIG. 9 illustrates a workflow of generating a custom data set for an artificial intelligence decision reapplication model, in accordance with some embodiments of the present disclosure;

[0018] FIG. 10 illustrates finetuning of a large language model when generating the artificial intelligence reapplication model, in accordance with some embodiments of the present disclosure;Attorney Docket No. 201818-0585628

[0019] FIGs. 11 A and 1 IB illustrate artificial intelligence decision reapplication model training and finetuning and results, in accordance with some embodiments of the present disclosure; and

[0020] FIG. 12 is a block diagram of an example of a computing system with which the present techniques may be implemented, in accordance with some embodiments of the present disclosure.

[0021] While the present techniques are susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. The drawings may not be to scale. It should be understood, however, that the drawings and detailed description thereto are not intended to limit the present techniques to the particular form disclosed, but to the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present techniques as defined by the appended claims.DETAILED DESCRIPTION OF CERTAIN EMBODIMENTS

[0022] To mitigate the problems described herein, the inventors had to both invent solutions and, in some cases just as importantly, recognize problems overlooked (or not yet foreseen) by others in the fields of machine learning, natural language processing, electronic messaging, and computer science. Indeed, the inventors wish to emphasize the difficulty of recognizing those problems that arc nascent and will become much more apparent in the future should trends in industry continue as the inventors expect. Further, because multiple problems are addressed, it should be understood that some embodiments are problem- specific, and not all embodiments address every problem with traditional systems described herein or provide every benefit described herein. That said, improvements that solve various permutations of these problems are described below.

[0023] As discussed above, Standards in messages are often used to provide a common understanding of the data across linguistic and system boundaries and to permit seamless, automated transmission, receipt, and processing of communications exchanged between institutions. Structured formats enable data to be unambiguous and machine friendly. However, many message systems necessarily include fields that accept free form or unstructured text to allow flexibility and capture important business context. The presence of such unstructured data often introduces ambiguity and processing errors, which create inefficiencies in automated workflows.Attorney Docket No. 201818-0585628

[0024] For example, the financial consequences of these inefficiencies are substantial. The global financial industry spends approximately $17 to $24 billion annually on trade processing functions and loses approximately $3 billion per year due to settlement inefficiencies. In post-trade settlement, low straight-through-processing (“STP”) rates can require extensive manual intervention by operations teams, delaying settlement and increasing operational and regulatory risk. These problems are magnified by shortened settlement cycles, such as the move to T+l and anticipated T+0 settlement regimes, and by settlement discipline rules that impose penalties for late or failed settlements.

[0025] Certain financial messages include information about conducting one or more financial transactions. Examples of such messages may be those covered by ISO 15022 MT540-3. For example, certain messages arc generated when funds are transferred internationally using the Society for Worldwide Interbank Financial Telecommunication (SWIFT) international payment network. These messages may include a variety of structured and / or unstructured data. For instance, messages may include one or more fields where only certain information can be provided, while other fields may allow free text or narrative. With respect to a SWIFT message, field 70E supports free text. As such, narratives regarding a transaction or settlement of a transaction may be included in this field. However, the inclusion of narrative in field 70E can lead to a significant proportion of non-STP trade settlement instructions. For example, instructions received from clients having a narrative message in a free text field, such as 70E, may cause the instruction to fail STP in a messaging system, such as a custody platform. A few examples of narratives that may be included in field 70E from incoming SWIFT messages include, SPRO / / PLEASE SUPPRESS FX SI, SPRO / / Instruct 22F field as OWNI, or SPRO / / NEW STOCK BORROW RETURN.

[0026] The narrative information can be manually crafted or system generated. The variety of information that can be included in a free text field presents many variations which typically requires special manual handling of incoming instructions. As such, a user who is processing instructions must make a decision as to whether to retain the narrative data on the outbound message to the market, strip it off and only retain internally, map the content of the narrative to a regular structured field on the outbound message, or other actions.Attorney Docket No. 201818-0585628

[0027] To address these challenges, embodiments of the present disclosure provide an artificial intclligcncc-bascd decision rcapplication engine configured to interpret narrative data and predict appropriate actions. A large language model (LLM) may be trained on historical inbound and outbound messages, including actions previously taken by operators in response to narratives. Based on this training, the model predicts an appropriate action for new messages, such as retaining, stripping, or amending structured fields. Predictions may be presented to a human operator for verification, or, when the model’s confidence exceeds a defined threshold, executed automatically.

[0028] In some embodiments, the model is trained and fine-tuned using Low-Rank Adaptation (LoRA) techniques. Training data is compiled by pairing inbound MT540-3 messages containing 70E narratives with corresponding outbound messages (with and without narratives) and linking them through reference numbers (e.g., the 20C SEME field). This pairing enables the system to learn historical decision outcomes at scale.

[0029] Training may proceed through a multi-phase process. First, inbound and outbound messages may be collected, curated, and converted into a machine-readable format such as JSON. Second, a high-parameter LLM may be used to generate reasoning explaining why specific actions were taken historically, thereby enriching the dataset. Third, a smaller LLM may be fine-tuned with LoRA adapters using the enriched dataset, producing a resource-efficient model capable of real-time inference. Finally, the fine-tuned model may be deployed to process live messages containing narrative.

[0030] When a new message with a narrative is received, the fine-tuned model may generate a predicted action and associated reasoning. If the prediction meets or exceeds a confidence threshold, the action may be executed automatically; otherwise, the message may be flagged for operator review. Testing of prototype systems demonstrated high performance, with structural and value prediction accuracy exceeding 98% and action prediction accuracy exceeding 97%.

[0031] Thus, the systems and methods of the present disclosure may provide insights that in some instances users cannot communicate via structured message fields and their struggle to instruct within existing field construct. The systems and methods, further presents a scalable approach which can be applied to other processes and message types to handle error and exception flowsAttorney Docket No. 201818-0585628 which involve interpreting free form text and requires manual intervention and may reduce the manual effort and risk of post trade processing thereby increasing speed and operational excellence while bringing down average unit cost, improve overall STP rate in the SWIFT messaging example, reduce time taken to process free format narrative elements allowing quicker transmission of messages thereby enabling more time for matching to be performed and issues to be identified / resolved, and other technical benefits to messaging systems that would be apparent to one of skill in the art in possession of the present disclosure.

[0032] FIG. 1 depicts a block diagram of an example of a messaging system 100, consistent with some embodiments. In some embodiments, the messaging system 100 may include a user computing device 102, a user computing device 103, a message processing computing device 104, and a messaging gateway 106. In various embodiments, the messaging system 100 may include a SWIFT messaging system. The user computing device 102 and the message processing computing device 104 may be in communication with each other over a network 108. In various embodiments, the user computing device 102 may be associated with a user (e.g., in memory of the messaging system 100 in virtue of user profiles). These various components may be implemented with computing devices like that shown in FIG. 8.

[0033] In some embodiments, the user computing device 102 or the user computing device 103 may be implemented using various combinations of hardware or software configured for wired or wireless communication over the network 108. For example, the user computing devices 102 and 103 may be implemented as a wireless telephone (e.g., small phone), a tablet, a personal digital assistant (PDA), a notebook computer, a personal computer, a connected set-top box (STB) such as provided by cable or satellite content providers, or a video game system console, a headmounted display (HIVID), a watch, an eyeglass projection screen, an autonomous / semi- autonomous device, a vehicle, a user badge, or other user computing devices. In some embodiments, the user computing devices 102 and 103 may include various combinations of hardware or software having one or more processors and capable of reading instructions stored on a tangible non-transitory machine-readable medium for execution by the one or more processors. Consistent with some embodiments, the user computing devices 102 and 103 include a machine- readable medium, such as a memory that includes instructions for execution by one or more processors for causing the user computing devices 102 and 103 to perform specific tasks. In someAttorney Docket No. 201818-0585628 embodiments, the instructions may be executed by the one or more processors in response to interaction by the user. In FIG. 1, two user computing devices arc shown, but commercial implementations are expected to include hundreds, thousands, or than one million, e.g., more than 10 million, geographically distributed over North America or the world.

[0034] The user computing devices 102 and 103 may include a communication system having one or more transceivers to communicate with other user computing devices or the message processing computing device 104. Accordingly, and as disclosed in further detail below, the user computing devices 102 and 103 may be in communication with systems directly or indirectly. As used herein, the phrase “in communication,” and variants thereof, is not limited to direct communication or continuous communication and may include indirect communication through one or more intermediary components or selective communication at periodic or aperiodic intervals, as well as one-time events.

[0035] For example, the user computing devices 102 and 103 in the messaging system 100 of FIG. 1 may include first (e.g., relatively long-range) transceiver to permit the user computing device 102 to communicate with the network 108 via a communication channel. In various embodiments, the network 108 may be implemented as a single network or a combination of multiple networks. For example, in various embodiments, the network 108 may include the Internet or one or more intranets, landline networks, wireless networks, or other appropriate types of communication networks. In another example, the network 108 may comprise a wireless telecommunications network adapted to communicate with other communication networks, such as the Internet. The wireless telecommunications network may be implemented by an example mobile cellular network, such as a long-term evolution (LTE) network or other third generation (3G), fourth generation (4G) wireless network, fifth generation (5G) wireless network or any subsequent generations. In some examples, the network 108 may be additionally or alternatively be implemented by a variety of communication networks, such as, but not limited to (which is not to suggest that other lists are limiting), a satellite communication network, a microwave radio network, or other communication networks.

[0036] The user computing devices 102 and 103 additionally may include second (e.g., short-range relative to the range of the first transceiver) transceiver to permit the user computing device 102Attorney Docket No. 201818-0585628 to communicate with each other or other user computing devices via a direct communication channel. Such second transceivers may be implemented by a type of transceiver supporting short- range (i.e., operate at distances that are shorter than the long-range transceivers) wireless networking. For example, such second transceivers may be implemented by Wi-Fi transceivers (e.g., via a Wi-Fi Direct protocol), Bluetooth® transceivers, infrared (IR) transceivers, and other transceivers that are configured to allow the user computing device 102 to communicate with each other or other user computing devices via an ad-hoc or other wireless network.

[0037] The messaging system 100 may also include or may be in connection with the message processing computing device 104. For example, the message processing computing device 104 may include one or more server devices, storage systems, cloud computing systems, or other computing devices (e.g., desktop computing device, laptop / notebook computing device, tablet computing device, mobile phone, etc.). In various embodiments, message processing computing device 104 may also include various combinations of hardware or software having one or more processors and capable of reading instructions stored on a tangible non-transitory machine- readable medium for execution by the one or more processors. Consistent with some embodiments, the message processing computing device 104 includes a machine-readable medium, such as a memory (not shown) that includes instructions for execution by one or more processors (not shown) for causing the message processing computing device 104 to perform specific tasks. In some embodiments, the instructions may be executed by the one or more processors in response to interaction by the user. The message processing computing device 104 may also be maintained by an entity with which sensitive credentials and information may be exchanged with the user computing devices 102 and 103. The message processing computing device 104 may further be one or more servers that hosts applications for the user computing device 102. The message processing computing device 104 may be more generally a messaging broker or a custody platform that processes messages transmitted by the messaging system between senders and one or more recipients.

[0038] The message processing computing device 104 may include various applications and may also be in communication with one or more external databases, that may provide additional information or data objects that may be used by the message processing computing device 104. For example, the message processing computing device 104 may obtain and provide messages,Attorney Docket No. 201818-0585628 via the network 108, messages from a messaging gateway 106 that may obtain or send messages that include unstructured data for the message processing computing device 104. While a specific messaging system 100 is illustrated in Fig. 1, one of skill in the art in possession of the present disclosure will recognize that other components and configurations are possible, and thus will fall under the scope of the present disclosure.

[0039] FIG. 2 illustrates an embodiment of a user computing device 200 that may be the user computing device 102 discussed above with reference to FIG. 1. In the illustrated embodiment, the user computing device 200 includes a chassis 202 that houses the components of the user computing device 200. Several of these components are illustrated in FIG. 2. For example, the chassis 202 may house a processing system and a non-transitory memory system that includes instructions that, when executed by the processing system, cause the processing system to provide a message application controller 204 that is configured to perform the functions of the message application controller or the user computing devices, discussed below. In the specific example illustrated in FIG. 2, the message application controller 204 is configured to provide one or more of a web browser application 204a or a native application 204b.

[0040] The chassis 202 may further house a communication system 210 that is coupled to the message application controller 204 (e.g., via a coupling between the communication system 210 and the processing system). The communication system 210 may include software or instructions that are stored on a computer-readable medium and that allow the user computing device 200 to send and receive messages through the communication networks discussed above. For example, the communication system 210 may include a communication interface to provide for communications through the network 108 as detailed above (e.g., first (e.g., long-range) transceiver). In an embodiment, the communication interface may include a wireless antenna that is configured to provide communications with IEEE 802.11 protocols (Wi-Fi), cellular communications, satellite communications, other microwave radio communications or communications. The communication system 210 may also include a communication interface (e.g., the second (e.g., short-range) transceiver) that is configured to provide direct communication with other user computing devices, sensors, storage devices, beacons, and other devices included in the securitization system discussed above with respect to FIG. 1. For example, the communication interface may include a wireless antenna that configured to operate accordingAttorney Docket No. 201818-0585628 to wireless protocols such as Bluetooth®, Bluetooth® Low Energy (BLE), near field communication (NFC), infrared data association (IrDA), ANT®, Zigbcc®, Z-Wavc® IEEE 802.11 protocols (Wi-Fi), or other wireless communication protocols that allow for direct communication between devices.

[0041] The chassis 202 may house a storage device (not illustrated) that provides a storage system 216 that is coupled to the message application controller 204 through the processing system. The storage system 216 may be configured to store data, applications, messages, or instructions described in further detail below and used to perform the functions described herein. In various embodiments, the chassis 202 also houses a user input / output (I / O) system 218 that is coupled to the message application controller 204 (e.g., via a coupling between the processing system and the user I / O system 218). In an embodiment, the user VO system 218 may be provided by a keyboard input subsystem, a mouse input subsystem, a track pad input subsystem, a touch input display subsystem, a microphone, an audio system, a haptic feedback system, or any other input subsystem. The chassis 202 also houses a display system 220 that is coupled to the message application controller 204 (e.g., via a coupling between the processing system and the display system 220) and may be included in the user VO system 218. In some embodiments, the display system 220 may be provided by a display device that is integrated into the user computing device 200 and that includes a display screen (e.g., a display screen on a laptop / notebook computing device, a tablet computing device, a mobile phone, or wearable device), or by a display device that is coupled directly to the user computing device 200 (e.g., a display device coupled to a desktop computing device by a cabled or wireless connection).

[0042] FIG. 3 depicts an embodiment of a message processing computing device 300, which may be the message processing computing device 104 discussed above with reference to FIG. 1. In the illustrated embodiment, the message processing computing device 300 includes a chassis 302 that houses the components of the message processing computing device 300, only some of which are illustrated in FIG. 3. For example, the chassis 302 may house a processing system (not illustrated) and a non-transitory memory system (not illustrated) that includes instructions that, when executed by the processing system, cause the processing system to provide a message decision reapplication engine 304 that is configured to perform the functions of the message decision reapplication engine or message processing computing device discussed below. Specifically, the message decisionAttorney Docket No. 201818-0585628 reapplication engine 304 may process messages that include fields that have unstructured data such as free form text as discussed in further detail below. The message decision rcapplication engine304 may be configured to receive and provide messages of the messaging system 100 over the network 108 to the message application controller included on the user computing device 102 / 200. For example, the user of the user computing device 102 / 200 may send or receive messages with the message decision reapplication engine 304 through the message application controller 204 over the network 108 to send or receive message communications or otherwise interact with the message decision reapplication engine 304.

[0043] The processing system and the non-transitory memory system may also include instructions that, when executed by the processing system, cause the processing system to provide an Al decision reapplication model 305 that is configured to perform the functions of the Al decision reapplication model or message processing computing device discussed below. For example, the Al decision reapplication model 305 may interpret unstructured data in messages and determine an action to be performed for the message or the unstructured data using various machine learning algorithms and artificial intelligence, as discussed in further detail below. The processing system and the non-transitory memory system may also include instructions that, when executed by the processing system, cause the processing system to provide an Al decision reapplication model generator 307 that may train or fine-tune the Al decision reapplication model305 using various training message data 310.

[0044] The chassis 302 may further house a communication system 306 that is coupled to the message decision reapplication engine 304 (e.g., via a coupling between the communication system 306 and the processing system) and that is configured to provide for communication through the network 108 of FIG. 1 as detailed below. The communication system 306 may allow the message processing computing device 300 to send and receive messages over the network 108 of FIG. 1. The chassis 302 may also house a storage device (not illustrated) that provides a storage system 308 that is coupled to the message decision reapplication engine 304through the processing system. The storage system 308 may be configured to store training message data 310 such as historical message data 310a that includes historical messages provided via the messaging system 100 of FIG. 1 and historical action data 310a taken with the historical message data. The storage system 308 may include other data or instructions to complete theAttorney Docket No. 201818-0585628 functionality discussed herein. In various embodiments, the storage system 308 may be provided on the message processing computing device 300 or on a database accessible via the communication system 306. Furthermore, while the message decision reapplication engine 304 is illustrated as being located on the message processing computing device 104 / 300, the message decision reapplication engine 304 may be included on the messaging gateway 106 of FIG. 1. For example, the message decision reapplication engine 304 may obtain and send a message or a portion of the messages via the messaging gateway 106 rather than receiving and sending the messages directly.

[0045] FIG. 4 illustrates an embodiment of a messaging gateway 400 that may be the messaging gateway 106 discussed above with reference to FIG. 1. In the illustrated embodiment, the messaging gateway 400 includes a chassis 402 that houses the components of the messaging gateway 400. Several of these components are illustrated in FIG. 4. For example, the chassis 402 may house a processing system and a non-transitory memory system that includes instructions that, when executed by the processing system, cause the processing system to provide a messaging gateway controller 404 that is configured to perform the functions of the messaging gateway controller or the messaging gateway, discussed below.

[0046] The chassis 402 may further house a communication system 406 that is coupled to the messaging gateway controller 404 (e.g., via a coupling between the communication system 406 and the processing system). The communication system 406 may include software or instructions that are stored on a computer-readable medium and that allow the messaging gateway 400 to send and receive messages through the communication networks discussed above. For example, the communication system 406 may include a communication interface to provide for communications through the network 108 as detailed above.

[0047] The chassis 402 may house a storage device (not illustrated) that provides a storage system 408 that is coupled to the messaging gateway controller 404 through the processing system. The storage system 408 may be configured to store data, applications, messages, or instructions described in further detail below and used to perform the functions described herein. For example, the storage system 408 may store historical messages that are outbound or inbound as well as current messages. While a specific example of a messaging gateway is illustrated, one of skill inAttorney Docket No. 201818-0585628 the art in possession of the present disclosure will recognize that different configurations may be contemplated.

[0048] FIG. 5 depicts an embodiment of a method 500 of message decision reapplication, which in some embodiments may be implemented with the components of FIGS. 1-4 discussed above. The method 500 illustrates how unstructured or partially unstructured fields of electronic messages may be interpreted, and how automated decisions can be applied to the unstructured data and the message as a whole. Examples of actions include, but are not limited to: (i) retaining the unstructured data and transmitting it to one or more recipients; (ii) stripping the unstructured data from the outbound message while retaining it in internal storage; or (iii) mapping or amending another field of the message based on the unstructured data (e.g., inserting content into a structured field). One of skill in the art will recognize that these embodiments improve existing messaging workflows by reducing the need for manual review and by increasing the consistency of message handling.

[0049] The method 500 is described as being performed primarily by the Al decision reapplication engine 304 included on the message processing computing device 104 / 300. In an embodiment, the method steps may additionally or alternatively be performed by the user computing device 102 / 200 or the messaging gateway 106 / 400. In some embodiments, the method may be distributed across multiple servers or processors, and may involve both local and cloud-based computation, such that the message parsing, interpretation, and action execution may occur on different nodes of the system.

[0050] The method 500 may proceed to block 502 where a message that is destined to one or more recipients and that includes a plurality of data fields, where a first set of the data fields includes unstructured data, is obtained. In an embodiment, at block 502, the message processing computing device 104 / 300 may receive a message generated by a user computing device 102 / 200 and transmitted via the messaging gateway 106 / 400, the message including one or more structured fields and one or more unstructured fields such as free form text in field 70E of a SWIFT ISO 15022 MT540-3 message. In some embodiments, block 502 may further include the message processing computing device 104 / 300 validating the message format, classifying the message type,Attorney Docket No. 201818-0585628 extracting metadata for processing, and filtering the message to identify whether it contains narrative content in designated fields, such as field 70E, prior to further analysis.

[0051] The method 500 may proceed to block 504 where the unstructured data in at least a portion of the first set of data fields is parsed to obtain parsed data. In an embodiment, at block 504, the message decision reapplication engine 304 may identify the unstructured data and converts the free form text into parsed data, for example through tokenization, normalization, or transformation into a machine-readable format such as JSON. In some embodiments, the parsing step further includes entity recognition or pattern matching to identify candidate values embedded in the text, such as account numbers, settlement references, legal entity identifiers (LEIs), CFETS trade reference numbers, or clearer details, which can then be isolated for interpretation by the Al model.

[0052] The method 500 may proceed to block 506 where the parsed data is processed with a large language model to determine an interpretation of the unstructured data and a confidence score in that interpretation. In an embodiment, at block 506, the Al decision reapplication model 305 processes the parsed data and generates an interpretation, such as “suppress FX standing instruction,” “map value to field 22F::SETR,” or “assign PSAF value to field 94F::SAFE,” together with a confidence score indicating the likelihood of correctness. In some embodiments, the interpretation step may include a multi-model approach, where reasoning generated by a high- parameter large language model is incorporated into the training dataset and used to enrich the context for a smaller fine-tuned model that performs inference in real time. In such cases, the smaller model may be fine-tuned using Low-Rank Adaptation (LoRA) techniques, thereby providing high accuracy at reduced computational cost. The Al decision reapplication model 305 may output multiple candidate interpretations ranked according to their confidence scores.

[0053] The method 500 may proceed to block 508 where, based on the interpretation and the confidence score, one or more actions are determined to take with at least one of the message or the unstructured data in the first set of data fields. In an embodiment, at block 508, the Al decision reapplication model 305 selects one or more actions, such as routing the message to an agent computing device for verification when the confidence score does not satisfy a first threshold, or automatically performing the actions when the confidence score satisfies a second, higher threshold. In some embodiments, the one or more actions may include: (i) retaining the narrativeAttorney Docket No. 201818-0585628 data in the outbound message; (ii) stripping the narrative data and storing it internally with a pointer to the message identifier; (iii) mapping the narrative into a structured field, such as amending field 22F::SETR, inserting a PSAF code into field 94F::SAFE, or adding clearer details into field 95P::DEAG; or (iv) performing a hybrid action, such as partially retaining the narrative while also updating a structured field.

[0054] The method 500 may proceed to block 510 where the one or more actions are performed with the message or the unstructured data. In an embodiment, at block 510, the message decision reapplication engine 304 modifies the message, updates one or more structured or unstructured fields, stores the action taken in a database together with a reference identifier for the message, and generates an audit log of the decision for compliance monitoring.

[0055] The method 500 then proceeds at block 512 by sending the message to the one or more recipients. In an embodiment, at block 512, the message decision reapplication engine 304 provides the message, after the one or more actions have been performed, to the messaging gateway 106 / 400 for onward delivery to a custody platform, a local custodian, or another recipient system. In some embodiments, the system may further feed the performed actions and resulting message back into a training dataset, along with agent feedback where applicable, to improve subsequent model performance and enable continuous learning of the Al decision reapplication model.

[0056] FIG. 6 depicts an embodiment of a method 600 of Al decision reapplication model training, which in some embodiments may be implemented with the components of FIGS. 1, 2, 3, and 4 discussed above. The method 600 may begin at block 602 where a plurality of historical messages that were sent via a messaging system and that each include a plurality of data fields, where a first set of data fields of the plurality of data fields includes unstructured data, are obtained. In an embodiment, at block 602, the Al decision reapplication model generator 307 obtains historical messages of the same type as those processed by the messaging system 100, such as inbound and outbound SWIFT ISO 15022 MT540-3 messages containing field 70E narrative content. In some embodiments, the historical messages may be retrieved from the messaging gateway 106 / 400 (e.g., historical messages 408a) and stored as historical message data 310a in the message processing computing device 300. In other embodiments, the messages may be curated into datasets by pairingAttorney Docket No. 201818-0585628 inbound messages containing narratives with corresponding outbound messages, linked by a reference identifier such as field 20C SEME, and converted into a structured format such as JSON for consistent training input.

[0057] The method 600 may proceed to block 604 where one or more historical actions taken for a respective historical message of the plurality of historical messages is obtained. In an embodiment, at block 604, the Al decision reapplication model generator 307 obtains historical action data 310b associated with the historical message data 310a. In some embodiments, the historical action data may be provided by a custody platform or other message processing system and may include whether narratives were retained, stripped, or mapped into structured fields. In other embodiments, historical action data may further capture operator annotations, repair queue decisions, or automated rules applied to previous trades.

[0058] The method 600 may proceed to block 606 where a large language model is trained using the one or more historical actions and at least a portion of the unstructured data in at least a portion of the first set of data fields of the plurality data fields in the historical messages. In an embodiment, at block 606, the Al decision reapplication model 305 is trained with the historical message data 310a and the historical action data 310b to learn correlations between free text narratives and the appropriate actions. In some embodiments, the training dataset may be enriched by reasoning generated by a high-parameter large language model, which explains why specific actions were taken historically, and this enriched dataset may then be used to fine-tune a smaller model using Low-Rank Adaptation (LoRA) techniques. In other embodiments, the training dataset may include agent feedback collected during execution of method 500 when unconfident decisions outputted by the Al decision reapplication model 305 are verified or rejected by a human operator. Randomly sampled confident predictions may also be reviewed to provide corrective feedback and improve accuracy.

[0059] The method 600 may proceed to block 608 where the large language model is stored, n an embodiment, at block 608, the trained or fine-tuned Al decision reapplication model 305 is stored in the storage system 308 for subsequent use during real-time inference. In some embodiments, the Al decision reapplication model 305 may be versioned, such that multiple iterations of the AlAttorney Docket No. 201818-0585628 decision reapplication model 305 are maintained, allowing rollback to prior versions or comparison of accuracy metrics over time.

[0060] The method 600 may proceed to block 610 where feedback from decisions produced by the large language model or additional historical messages are received. In an embodiment, at block 610, the Al decision reapplication model generator 307 may receive feedback from an agent of the custody program when unconfident decisions generated by the Al decision reapplication model 305 are approved or rejected by the agent. In other embodiments, feedback may be obtained from post- settlement outcomes, from anomalies detected in downstream systems, or from random sampling of high-confidence decisions. The additional historical messages may be continuously ingested to expand and diversify the training dataset.

[0061] In an embodiment, at block 612, the Al decision reapplication model 305 is retrained or fine-tuned with the new data and feedback. The process may iterate through block 608 such that the Al decision reapplication model 305 is continuously updated to improve performance, reduce error rates, and adapt to new narrative types or market conditions. In some embodiments, periodic evaluation metrics, such as structural accuracy and action prediction accuracy, may be recorded and used to determine when retraining is warranted or when a new model should be promoted to production.

[0062] Accordingly, the method 600 provides a framework for training and continuously refining the Al decision reapplication model 305. By leveraging historical message and action data, agent feedback, and enriched reasoning, the system improves its ability to accurately interpret unstructured narrative content and recommend appropriate actions. These iterative updates enable the model to adapt to evolving market practices and message formats, reduce manual repair effort, and increase overall straight-through -processing (STP) rates in post-trade workflows. One of skill in the art will recognize that while method 600 is described with respect to financial trade messages, the approach is broadly applicable to other domains in which unstructured data must be consistently interpreted and acted upon within structured messaging systems.

[0063] ] FIG. 7 illustrates a workflow of message decision reapplication during the method of FIG.5 and training of the message decision reapplication model during the method of FIG. 6, in accordance with some embodiments of the present disclosure. Process 702 illustrates a currentAttorney Docket No. 201818-0585628 state of a messaging system without Al assistance. At 702a, a trade instruction message, such as a SWIFT ISO 15022 MT540-3 message from a client, is transmitted to a messaging gateway 702b. The messaging gateway 702b provides the message to a custody platform 702c for processing. At 702d, if the message contains unstructured data such as a free text field (e.g., field 70E narrative), the message is routed to a repair queue 702e. At this stage, an agent manually determines whether to (i) retain the unstructured data in the outbound message and send it to the market, (ii) strip the unstructured data and store it internally, or (iii) use the unstructured data to amend another structured or unstructured field of the message. Once the agent makes the determination, the message is returned to the messaging gateway 702b and transmitted to one or more recipients at 702f. This workflow is resource-intensive, introduces delays, and may lead to inconsistent outcomes across different agents.

[0064] Process 704 illustrates embodiments of the present disclosure and the improved workflow using methods 500 and 600. At 704a, a message is sent to a messaging gateway 704b, which may correspond to the messaging gateway 106 / 400. The messaging gateway 704b provides the message to the message processing computing device 104 / 300 at 704c. At 704d, the message is processed by the Al decision reapplication model 305, which interprets unstructured data (e.g., free text field) and outputs a predicted decision and associated action to the custody platform 704e. In some embodiments, the custody platform 704e may apply the predicted action directly, while in other embodiments the action may be reviewed by an approver at 704f, providing a human-in-the-loop safeguard when the confidence score is below a threshold or when business rules require oversight. After execution or approval, the message is provided back to the messaging gateway 704b and transmitted to one or more recipients at 704g. By incorporating Al-based inference, process 704 reduces reliance on repair queues, accelerates message handling, and improves the consistency of outcomes.

[0065] In various embodiments, the Al decision reapplication model 305 may be trained or finetuned by the Al decision reapplication model generator 307 using multiple data sources. At 704h, historical messages may be obtained from the messaging gateway 704b, including inbound and outbound message pairs. At 704i, historical actions performed by agents in process 702, such as retaining, stripping, or amending narratives, may be incorporated into the training dataset. At 704j, feedback from approvers reviewing Al predictions may also be captured and used to further refineAttorney Docket No. 201818-0585628 the model. Together, these data sources enable the Al decision reapplication model generator 307 to iteratively update and improve the Al decision rcapplication model 305, allowing the system to adapt to new message formats, narrative types, and evolving business practices.

[0066] Accordingly, FIG. 7 demonstrates how the disclosed techniques improve upon traditional repair queue processing by replacing inconsistent and time-consuming manual interventions with an automated Al-assisted workflow. Process 702 illustrates the current state, where messages containing unstructured data are diverted to a repair' queue and require agent review, introducing latency and additional cost. In contrast, process 704 integrates the Al decision reapplication model 305 into the message flow, enabling real-time interpretation of narratives, application of predicted actions, and human-in-the-loop verification when needed. This approach reduces repair queue dependency, improves straight-through-processing (STP) rates, and supports more consistent handling of unstructured data. Moreover, the feedback loop involving historical messages, historical actions, and agent verification enables continuous model improvement, thereby enhancing system performance over time and adapting to new message formats and evolving business requirements.

[0067] FIG. 8 illustrates a workflow 800 of the method 500 of FIG. 5 and the method 600 of FIG.6, in accordance with some embodiments. The workflow 800 provides an end-to-end process for preparing training data, training large language models, and deploying those models for inference on live messages. In a first phase 802, historical message data is collected and curated. At 804, historical incoming messages may be obtained from custody records 806, while at 808, historical outgoing messages are also collected. At 810, the inbound and outbound messages are matched based on a common identifier, such as a trade reference number (e.g., field 20C SEME), to provide proper pairing of related messages. At 812, relevant fields are extracted from the matched messages, including fields that contain unstructured or free text content, such as narrative data from field 70E of SWIFT MT540-3 messages. In some embodiments, the messages or extracted data may be normalized and converted into a structured representation, such as JSON format, to facilitate consistency in training. The curated and normalized data is then stored as a training dataset at 814. This phase helps the training corpus accurately reflect the actions and decisions historically made with respect to narrative fields.Attorney Docket No. 201818-0585628

[0068] The workflow 800 may proceed to a second phase 816 in which reasoning is generated using a high-parameter large language model. At 818, the curated training dataset and, at 820, one or more prompts are provided as inputs to a large LLM at 821, which may be a general-purpose model with billions of parameters. At 822, the large LLM generates reasoning or explanations as to why changes occurred between inbound and outbound message pairs in the original dataset. For example, the reasoning may identify that a free text instruction in field 70E corresponded to a change in field 22F::SETR, or that an FX suppression note was removed before transmission to the market. At 824, an enhanced training dataset is created that incorporates both the original inbound / outbound message pairs and the generated reasoning. By incorporating explicit reasoning, the dataset provides richer context for training downstream models, allowing smaller models to better capture decision-making logic that would otherwise require significant manual annotation.

[0069] The workflow 800 may proceed to a third phase 826 in which a low-parameter model is fine-tuned. At 828, the enhanced training dataset from 824 and a prompt at 830 are provided as inputs to a smaller LLM model at 832. The smaller LLM, which may be parameterized for efficient inference, produces fine-tuned outputs at 835 that reflect the enriched reasoning. A Low-Rank Adaptation (LoRA) adapter 834 is trained using these outputs to provide lightweight, efficient fine-tuning of the smaller model. At 836, the LoRA adapter is merged with the base LLM to generate a fine-tuned model at 838. This approach allows the benefits of the reasoning generated by the larger model to be distilled into a smaller, more efficient model that is suitable for real-time inference in production environments, while minimizing computational costs.

[0070] The workflow 800 may proceed to a fourth phase 840 in which the fine-tuned model is deployed to generate predictions for live incoming messages. At 842, a new incoming message and, at 844, a prompt may be provided as inputs to the fine-tuned LLM model at 846. The model outputs a prediction at 848, which may include an action class such as “retain,” “strip,” or “amend,” along with a specific field-level modification. At 850, a prediction decision is made based on the model output, and at 852, an action is executed accordingly, such as modifying a structured field or retaining narrative data internally. In some embodiments, the predicted action is also displayed to a user computing device at 854 for review. At 856, the user interface presents the predicted actions to an operator, who may accept, reject, or edit the actions before the message is finalized and passed out of the system as an outgoing message. This human-in-the-loop safeguard providesAttorney Docket No. 201818-0585628 quality control while still achieving significant automation gains. Feedback from operator edits can further be incorporated into subsequent training iterations, creating a continuous learning loop.

[0071] Accordingly, FIG. 8 illustrates how the disclosed techniques provide an integrated pipeline for data curation, reasoning generation, fine-tuning, and deployment of artificial intelligence models for message decision reapplication. By pairing historical inbound and outbound messages, converting them into structured formats, and enriching the training dataset with reasoning from a high-parameter model, the system produces a fine-tuned and resource-efficient model that can be deployed in real-time messaging environments. This workflow enables automation of narrative interpretation and action selection, while still allowing for human oversight when appropriate. Moreover, by feeding operator feedback and new historical data back into the training loop, the system achieves continuous learning and adaptability. The result is a scalable and efficient approach that reduces manual intervention, improves STP rates, and provides consistent handling of unstructured data across diverse messaging scenarios.

[0072] FIG. 9 illustrates an example 900 of the second phase 816 of FIG. 8, in which building a custom dataset for training an Al decision reapplication model is performed. At 902, the data collection phase 802 of FIG. 8 is shown, where historical inbound and outbound messages are obtained from a custody platform and paired with one another. Each pair may represent an inbound message received from a client and the corresponding outbound message transmitted to the market, thereby capturing the historical transformation of trade instructions. In some embodiments, only pairs containing unstructured data fields, such as narrative fields (e.g., field 70E in a SWIFT MT540-3 message), are retained for further processing, such that the dataset is focused on instances where narrative-driven decision-making occurs.

[0073] The paired messages may be normalized and converted into a structured representation, such as JSON format, to provide consistency across the training corpus. In-scope fields are then extracted for each message pair, including both unstructured fields (e.g., narrative text) and any structured fields that were amended as a result of the narrative. At this stage, a logic flow may be applied to automatically label each message pair with an action class based on observed historical outcomes, such as “retain,” “strip,” “amend,” or “map to a different field.” The labeled datasetAttorney Docket No. 201818-0585628 generated through this process is referred to as intermediate data 904, which corresponds to the training dataset at 814 of FIG. 8.

[0074] The intermediate data 904, together with a carefully designed prompt 906, may then be provided as input to a large language model 908, such as the Mixtral 8x22B model. The large language model 908 generates reasoning 910 that explains why specific actions were historically taken in response to narrative fields. For example, the model may infer that a narrative instruction “SPRO / / OWNI” led to an amendment in field 22F::SETR, or that a free text FX suppression note was stripped before being transmitted to the market. The generated reasoning enriches the dataset by adding contextual explanations to the labeled actions.

[0075] The reasoning 910 may then be combined with the intermediate data 904 to form a final dataset 912, which represents the enhanced training dataset at 824 of FIG. 8. This final dataset not only captures the structural relationships between inbound and outbound messages and their labeled outcomes, but also embeds human-like reasoning that can be distilled into smaller models during fine-tuning. Accordingly, the process of FIG. 9 illustrates how augmenting labeled historical message pairs with generated reasoning produces an enhanced training dataset that improves both the interpretability and predictive power of Al decision reapplication models. By embedding contextual explanations alongside labeled outcomes, the system enables smaller finetuned models to achieve high accuracy with lower computational cost, thereby supporting realtime deployment while preserving consistency and reducing manual intervention in handling unstructured data.

[0076] FIG. 10 and FIGs. 11A and 11B illustrate training and fine-tuning of the model with the enhanced training dataset at 824 of FIG. 8. As shown in the table 1000 of FIG. 10, fine-tuning can significantly reduce the computational resources required for model training and deployment. For example, compared to full parameter training of a large language model, fine-tuning with LoRA adapters reduces memory usage for storing gradients and optimization states and lowers the overall GPU footprint. This resource efficiency allows large-scale training to be performed more quickly and at reduced cost, while enabling deployment of fine-tuned models on hardware with limited processing or memory capacity. In this way, the disclosed techniques address not only the accuracyAttorney Docket No. 201818-0585628 of Al predictions but also the practical considerations of scalability and implementation in production environments.

[0077] FIGs. 11A and 11B illustrate how the enhanced training dataset 912 of FIG. 9 may be incorporated into the fine-tuning process. At 1102, the dataset 912 is input to a large language model that has been pretrained on general-purpose corpora. The pretrained model parameters are then refined using the enhanced dataset through application of LoRA weights 1104. In particular, LoRA adapts only a low-rank subset of model parameters, leaving the majority of the pretrained weights unchanged, which enables efficient adaptation of the model to the specialized domain of financial message narratives without requiring the cost of retraining the entire network. The resulting fine-tuned LLM at 838 of FIG. 8 serves as the Al decision reapplication model 305, which can be used to process current incoming messages in real time. By applying the enriched reasoning and domain- specific labels from the enhanced dataset, the fine-tuned model improves prediction accuracy for unstructured data handling while maintaining computational efficiency.

[0078] Accordingly, FIGS. 10 and 11A and 11B illustrate how the disclosed training framework combines enriched domain- specific datasets with efficient fine-tuning strategies to generate Al models that are both accurate and practical for deployment. The use of LoRA adapters enables the system to leverage the strengths of large pretrained models while tailoring them to the unique challenges of narrative data in financial messages or other domain messages. This approach helps the Al decision reapplication model 305 to deliver high-confidence, real-time predictions in production systems while minimizing resource consumption and training overhead.

[0079] Thus, the systems and methods of the present disclosure provide a comprehensive framework for collecting and curating message data, generating enriched reasoning datasets, fine- tuning large language models with efficient adaptation techniques, and deploying those models for real-time inference in production environments. Unlike conventional systems that require extensive manual intervention or resource-intensive training of full models, the disclosed techniques improve the functioning of computer systems and artificial intelligence engines themselves. For example, the use of Low-Rank Adaptation (LoRA) significantly reduces GPU memory usage and training overhead, enabling deployment of accurate models on hardware with limited processing resources. Similarly, the enrichment of training data with generated reasoningAttorney Docket No. 201818-0585628 allows smaller, fine-tuned models to capture complex decision-making logic that would otherwise demand larger models or manual rule coding. Together, these improvements increase the technical efficiency of message processing systems by reducing computational cost, latency, and error rates, while enabling consistent handling of unstructured narrative data. The disclosed techniques therefore enhance both the performance of the underlying computer infrastructure and the accuracy of Al-driven decision engines, delivering measurable improvements in throughput, scalability, and reliability in real-world messaging environments.

[0080] FIG. 12 is a diagram that illustrates an exemplary computing system 1200 in accordance with embodiments of the present technique. The user computing device 102, 103, and 200, the message processing computing devices 104 and 300, and the messaging gateway 106 and 400, discussed above, may be provided by the computing system 1200. Various portions of systems and methods described herein, may include or be executed on one or more computing systems similar to computing system 1200. Further, processes and modules described herein may be executed by one or more processing systems similar to that of computing system 1200.

[0081] Computing system 1200 may include one or more processors (e.g., processors 1210a- 1210n) coupled to system memory 1220, an input / output I / O device interface 1230, and a network interface 1240 via an input / output (VO) interface 1250. A processor may include a single processor or a plurality of processors (e.g., distributed processors). A processor may be any suitable processor capable of executing or otherwise performing instructions. A processor may include a central processing unit (CPU) that carries out program instructions to perform the arithmetical, logical, and input / output operations of computing system 1200. A processor may execute code (e.g., processor firmware, a protocol stack, a database management system, an operating system, or a combination thereof) that creates an execution environment for program instructions. A processor may include a programmable processor. A processor may include general or special purpose microprocessors. A processor may receive instructions and data from a memory (e.g., system memory 1220). Computing system 1200 may be a uni-processor system including one processor (e.g., processor 1210a), or a multi-processor system including any number of suitable processors (e.g., 1210a-1210n). Multiple processors may be employed to provide for parallel or sequential execution of one or more portions of the techniques described herein. Processes, such as logic flows, described herein may be performed by one or more programmable processorsAttorney Docket No. 201818-0585628 executing one or more computer programs to perform functions by operating on input data and generating corresponding output. Processes described herein may be performed by, and apparatus may also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit). Computing system 1200 may include a plurality of computing devices (e.g., distributed computing systems) to implement various processing functions.

[0082] I / O device interface 1230 may provide an interface for connection of one or more I / O devices 1260 to computing system 1200. I / O devices may include devices that receive input (e.g., from a user) or output information (e.g., to a user). VO devices 1260 may include, for example, graphical user interface presented on displays (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor), pointing devices (e.g., a computer mouse or trackball), keyboards, keypads, touchpads, scanning devices, voice recognition devices, gesture recognition devices, printers, audio speakers, microphones, cameras, or the like. VO devices 1260 may be connected to computing system 1200 through a wired or wireless connection. VO devices 1260 may be connected to computing system 1200 from a remote location. VO devices 1260 located on remote computing system, for example, may be connected to computing system 1200 via a network and network interface 1240.

[0083] Network interface 1240 may include a network adapter that provides for connection of computing system 1200 to a network. Network interface 1240 may facilitate data exchange between computing system 1200 and other devices connected to the network. Network interface 1240 may support wired or wireless communication. The network may include an electronic communication network, such as the Internet, a local area network (LAN), a wide area network (WAN), a cellular communications network, or the like.

[0084] System memory 1220 may be configured to store program instructions 1201 or data 1202. Program instructions 1201 may be executable by a processor (e.g., one or more of processors 1210a-1210n) to implement one or more embodiments of the present techniques. Instructions 1201 may include modules of computer program instructions for implementing one or more techniques described herein with regard to various processing modules. Program instructions may include a computer program (which in certain forms is known as a program, software, software application,Attorney Docket No. 201818-0585628 script, or code). A computer program may be written in a programming language, including compiled or interpreted languages, or declarative or procedural languages. A computer program may include a unit suitable for use in a computing environment, including as a stand-alone program, a module, a component, or a subroutine. A computer program may or may not correspond to a file in a file system. A program may be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub programs, or portions of code). A computer program may be deployed to be executed on one or more computer processors located locally at one site or distributed across multiple remote sites and interconnected by a communication network.

[0085] System memory 1220 may include a tangible program carrier having program instructions stored thereon. A tangible program carrier may include a non-transitory computer readable storage medium. A non-transitory computer readable storage medium may include a machine readable storage device, a machine readable storage substrate, a memory device, or any combination thereof. Non-transitory computer readable storage medium may include non-volatile memory (e.g., flash memory, ROM, PROM, EPROM, EEPROM memory), volatile memory (e.g., random access memory (RAM), static random access memory (SRAM), synchronous dynamic RAM (SDRAM)), bulk storage memory (e.g., CD-ROM or DVD-ROM, hard-drives), or the like. System memory 1220 may include a non-transitory computer readable storage medium that may have program instructions stored thereon that are executable by a computer processor (e.g., one or more of processors 1210a-610n) to cause the subject matter and the functional operations described herein. A memory (e.g., system memory 1220) may include a single memory device or a plurality of memory devices (e.g., distributed memory devices). Instructions or other program code to provide the functionality described herein may be stored on a tangible, non-transitory computer readable media. In some cases, the entire set of instructions may be stored concurrently on the media, or in some cases, different parts of the instructions may be stored on the same media at different times.

[0086] VO interface 1250 may be configured to coordinate VO traffic between processors 1210a- 1210n, system memory 1220, network interface 1240, VO devices 1260, or other peripheral devices. VO interface 1250 may perform protocol, timing, or other data transformations to convertAttorney Docket No. 201818-0585628 data signals from one component (e.g., system memory 1220) into a format suitable for use by another component (e.g., processors 1210a- 1210n). I / O interface 1250 may include support for devices attached through various types of peripheral buses, such as a variant of the Peripheral Component Interconnect (PCI) bus standard or the Universal Serial Bus (USB) standard.

[0087] Embodiments of the techniques described herein may be implemented using a single instance of computing system 1200 or multiple computing systems 1200 configured to host different portions or instances of embodiments. Multiple computing systems 1200 may provide for parallel or sequential processing / execution of one or more portions of the techniques described herein.

[0088] Those skilled in the art will appreciate that computing system 1200 is merely illustrative and is not intended to limit the scope of the techniques described herein. Computing system 1200 may include any combination of devices or software that may perform or otherwise provide for the performance of the techniques described herein. For example, computing system 1200 may include or be a combination of a cloud-computing system, a data center, a server rack, a server, a virtual server, a desktop computer, a laptop computer, a tablet computer, a server device, a client device, a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a vehicle-mounted computer, or a Global Positioning System (GPS), or the like. Computing system 1200 may also be connected to other devices that are not illustrated, or may operate as a stand-alone system. In addition, the functionality provided by the illustrated components may in some embodiments be combined in fewer components or distributed in additional components. Similarly, in some embodiments, the functionality of some of the illustrated components may not be provided or other additional functionality may be available.

[0089] Those skilled in the art will also appreciate that while various items are illustrated as being stored in memory or on storage while being used, these items or portions of them may be transferred between memory and other storage devices for purposes of memory management and data integrity. Alternatively, in other embodiments some or all of the software components may execute in memory on another device and communicate with the illustrated computing system via inter-computer communication. Some or all of the system components or data structures may also be stored (e.g., as instructions or structured data) on a computer-accessible medium or a portableAttorney Docket No. 201818-0585628 article to be read by an appropriate drive, various examples of which are described above. In some embodiments, instructions stored on a computer-accessible medium separate from computing system 1200 may be transmitted to computing system 1200 via transmission media or signals such as electrical, electromagnetic, or digital signals, conveyed via a communication medium such as a network or a wireless link. Various embodiments may further include receiving, sending, or storing instructions or data implemented in accordance with the foregoing description upon a computer- accessible medium. Accordingly, the present techniques may be practiced with other computing system configurations.

[0090] In block diagrams, illustrated components are depicted as discrete functional blocks, but embodiments are not limited to systems in which the functionality described herein is organized as illustrated. The functionality provided by each of the components may be provided by software or hardware modules that are differently organized than is presently depicted, for example such software or hardware may be intermingled, conjoined, replicated, broken up, distributed (e.g. within a data center or geographically), or otherwise differently organized. The functionality described herein may be provided by one or more processors of one or more computers executing code stored on a tangible, non-transitory, machine readable medium. In some cases, notwithstanding use of the singular term "medium," the instructions may be distributed on different storage devices associated with different computing devices, for instance, with each computing device having a different subset of the instructions, an implementation consistent with usage of the singular term “medium” herein. In some cases, third party content delivery networks may host some or all of the information conveyed over networks, in which case, to the extent information (e.g., content) is said to be supplied or otherwise provided, the information may be provided by sending instructions to retrieve that information from a content delivery network.

[0091] The reader should appreciate that the present application describes several independently useful techniques. Rather than separating those techniques into multiple isolated patent applications, applicants have grouped these techniques into a single document because their related subject matter lends itself to economies in the application process. But the distinct advantages and aspects of such techniques should not be conflated. In some cases, embodiments address all of the deficiencies noted herein, but it should be understood that the techniques are independently useful, and some embodiments address only a subset of such problems or offer other, unmentionedAttorney Docket No. 201818-0585628 benefits that will be apparent to those of skill in the art reviewing the present disclosure. Due to costs constraints, some techniques disclosed herein may not be presently claimed and may be claimed in later filings, such as continuation applications or by amending the present claims. Similarly, due to space constraints, neither the Abstract nor the Summary of the Invention sections of the present document should be taken as containing a comprehensive listing of all such techniques or all aspects of such techniques.

[0092] It should be understood that the description and the drawings are not intended to limit the present techniques to the particular form disclosed, but to the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present techniques as defined by the appended claims. Further modifications and alternative embodiments of various aspects of the techniques will be apparent to those skilled in the art in view of this description. Accordingly, this description and the drawings are to be construed as illustrative only and are for the purpose of teaching those skilled in the ail the general manner of carrying out the present techniques. It is to be understood that the forms of the present techniques shown and described herein are to be taken as examples of embodiments. Elements and materials may be substituted for those illustrated and described herein, pails and processes may be reversed or omitted, and certain features of the present techniques may be utilized independently, all as would be apparent to one skilled in the art after having the benefit of this description of the present techniques. Changes may be made in the elements described herein without departing from the spirit and scope of the present techniques as described in the following claims. Headings used herein are for organizational purposes only and are not meant to be used to limit the scope of the description.

[0093] As used throughout this application, the word “may” is used in a permissive sense (i.e., meaning having the potential to), rather than the mandatory sense (i.e., meaning must). The words “include”, “including”, and “includes” and the like mean including, but not limited to. As used throughout this application, the singular forms “a,” “an,” and “the” include plural referents unless the content explicitly indicates otherwise. Thus, for example, reference to “an element” or "a element" includes a combination of two or more elements, notwithstanding use of other terms and phrases for one or more elements, such as “one or more.” The term "or" is, unless indicated otherwise, non-exclusive, i.e., encompassing both "and" and "or." Terms describing conditionalAttorney Docket No. 201818-0585628 relationships, e.g., "in response to X, Y," "upon X, Y,", “if X, Y,” "when X, Y," and the like, encompass causal relationships in which the antecedent is a necessary causal condition, the antecedent is a sufficient causal condition, or the antecedent is a contributory causal condition of the consequent, e.g., "state X occurs upon condition Y obtaining" is generic to "X occurs solely upon Y" and "X occurs upon Y and Z." Such conditional relationships are not limited to consequences that instantly follow the antecedent obtaining, as some consequences may be delayed, and in conditional statements, antecedents are connected to their consequents, e.g., the antecedent is relevant to the likelihood of the consequent occurring. Statements in which a plurality of attributes or functions are mapped to a plurality of objects (e.g., one or more processors performing steps A, B, C, and D) encompasses both all such attributes or functions being mapped to all such objects and subsets of the attributes or functions being mapped to subsets of the attributes or functions (e.g., both all processors each performing steps A-D, and a case in which processor 1 performs step A, processor 2 performs step B and part of step C, and processor 3 performs part of step C and step D), unless otherwise indicated. Similarly, reference to “a computing system” performing step A and “the computing system” performing step B can include the same computing device within the computing system performing both steps or different computing devices within the computing system performing steps A and B. Further, unless otherwise indicated, statements that one value or action is “based on” another condition or value encompass both instances in which the condition or value is the sole factor and instances in which the condition or value is one factor among a plurality of factors. Unless otherwise indicated, statements that “each” instance of some collection have some property should not be read to exclude cases where some otherwise identical or similar members of a larger collection do not have the property, i.e., each does not necessarily mean each and every. Limitations as to sequence of recited steps should not be read into the claims unless explicitly specified, e.g., with explicit language like “after performing X, performing Y,” in contrast to statements that might be improperly argued to imply sequence limitations, like “performing X on items, performing Y on the X’ed items,” used for purposes of making claims more readable rather than specifying sequence. Statements referring to “at least Z of A, B, and C,” and the like (e.g., “at least Z of A, B, or C”), refer to at least Z of the listed categories (A, B, and C) and do not require at least Z units in each category. Unless specifically stated otherwise, as apparent from the discussion, it is appreciated that throughout this specification discussions utilizing terms such as “processing,”Attorney Docket No. 201818-0585628“computing,” “calculating,” “determining” or the like refer to actions or processes of a specific apparatus, such as a special purpose computer or a similar special purpose electronic processing / computing device. Features described with reference to geometric constructs, like "parallel," "perpendicular / orthogonal," “square”, “cylindrical,” and the like, should be construed as encompassing items that substantially embody the properties of the geometric construct, e.g., reference to "parallel" surfaces encompasses substantially parallel surfaces. The permitted range of deviation from Platonic ideals of these geometric constructs is to be determined with reference to ranges in the specification, and where such ranges are not stated, with reference to industry norms in the field of use, and where such ranges are not defined, with reference to industry norms in the field of manufacturing of the designated feature, and where such ranges are not defined, features substantially embodying a geometric construct should be construed to include those features within 15% of the defining attributes of that geometric construct. The terms "first", "second", "third," “given” and so on, if used in the claims, are used to distinguish or otherwise identify, and not to show a sequential or numerical limitation. As is the case in ordinary usage in the field, data structures and formats described with reference to uses salient to a human need not be presented in a human-intelligible format to constitute the described data structure or format, e.g., text need not be rendered or even encoded in Unicode or ASCII to constitute text; images, maps, and data-visualizations need not be displayed or decoded to constitute images, maps, and data-visualizations, respectively; speech, music, and other audio need not be emitted through a speaker or decoded to constitute speech, music, or other audio, respectively. Computer implemented instructions, commands, and the like are not limited to executable code and can be implemented in the form of data that causes functionality to be invoked, e.g., in the form of arguments of a function or API call. To the extent bespoke noun phrases (and other coined terms) are used in the claims and lack a self-evident construction, the definition of such phrases may be recited in the claim itself, in which case, the use of such bespoke noun phrases should not be taken as invitation to impart additional limitations by looking to the specification or extrinsic evidence.

[0094] In this patent, to the extent any U.S. patents, U.S. patent applications, or other materials (e.g., articles) have been incorporated by reference, the text of such materials is only incorporated by reference to the extent that no conflict exists between such material and the statements and drawings set forth herein. In the event of such conflict, the text of the present document governs,Attorney Docket No. 201818-0585628 and terms in this document should not be given a narrower reading in virtue of the way in which those terms arc used in other materials incorporated by reference.

[0095] The present techniques will be better understood with reference to the following enumerated embodiments:1. A system, comprising: one or more processors; and memory storing instructions that when executed by the one or more processors cause the one or more processors to effectuate operations comprising: receiving a message that is destined to one or more recipients and that includes a plurality of data fields, wherein a first set of data fields of the plurality of data fields includes unstructured data; parsing the unstructured data in at least a portion of the first set of data fields to obtain parsed data; processing, with a large language model, the parsed data to determine an interpretation of the unstructured data and a confidence score in that interpretation; determining, based on the interpretation and the confidence score, one or more actions to take with at least one of the message or the unstructured data in the at least the portion of the first set of data fields; performing the one or more actions; and sending the message to the one or more recipients.2. The system of embodiment 1, wherein the one or more actions includes: sending to an agent computing device for verification of the interpretation and the one or more actions if the confidence score does not satisfy a threshold.3. The system of any one of embodiments 1 or 2, wherein the one or more actions includes: providing instructions to automatically perform the one or more actions if the confidence score satisfies a threshold.4. The system of any one of embodiments 1-3, wherein the one or more actions includes: retaining the unstructured data as the unstructured data was originally presented in the message.Attorney Docket No. 201818-05856285. The system of any one of embodiments 1-4, wherein the one or more actions includes: stripping the unstructured data from the message; and retaining the unstructured data in storage with an association to the message.6. The system of any one of embodiments 1-5, wherein the one or more actions includes: using the unstructured data to amend one or more data fields of the plurality of data fields where data in the one or more data fields includes structured data, semi-unstructured data, or unstructured data.7. The system of any one of embodiments 1-6, wherein the operations further comprise: obtaining a plurality of historical messages that were sent via a messaging system and that each include a plurality of data fields, wherein a set of data fields of the plurality of data fields includes unstructured data; obtaining one or more historical actions taken for a respective historical message of the plurality of historical messages; training, using the one or more historical actions and at least a portion of the unstructured data in at least a portion of the set of data fields of the plurality of data fields, the large language model; and saving the large language model in storage.8. The system of embodiment 7, wherein obtaining the plurality of historical messages includes pairing inbound messages of the plurality of historical messages that include the unstructured data with outbound messages of the plurality of historical messages, and linking the inbound messages and the outbound messages via a reference identifier included in each of the inbound messages and the outbound messages.9. The system of any one of embodiments 7 or 8, wherein training the large language model comprises fine-tuning the large language model using a Low-Rank Adaptation (LoRA) technique.10. The system of any one of embodiments 7-9, wherein the training the large language model further includes generating reasoning using a high-parameter language model and incorporating the reasoning into a training dataset for the large language model.Attorney Docket No. 201818-058562811. A non-transitory, machine-readable medium storing instructions that, when executed by one or more processors, effectuate operations comprising: receiving, by a computing system, a message that is destined to one or more recipients and that includes a plurality of data fields, wherein a first set of data fields of the plurality of data fields includes unstructured data; parsing, by the computing system, the unstructured data in at least a portion of the first set of data fields to obtain parsed data; processing, by the computing system and with a large language model, the parsed data to determine an interpretation of the unstructured data and a confidence score in that interpretation; determining, by the computing system and based on the interpretation and the confidence score, one or more actions to take with at least one of the message or the unstructured data in the at least the portion of the first set of data fields; performing, by the computing system, the one or more actions; and sending, by the computing system, the message to the one or more recipients.12. The non-transitory, machine-readable medium of embodiment 11, wherein the one or more actions includes: sending to an agent computing device for verification of the interpretation and the one or more actions if the confidence score does not satisfy a threshold.13. The non-transitory, machine-readable medium of any one of embodiments 11 or 12, wherein the one or more actions includes: providing instructions to automatically perform the one or more actions if the confidence score satisfies a threshold.14. The non-transitory, machine-readable medium of any one of embodiments 11-13, wherein the one or more actions includes: retaining the unstructured data as the unstructured data was originally presented in the message.15. The non-transitory, machine-readable medium of any one of embodiments 11-14, wherein the one or more actions includes: stripping the unstructured data from the message; and retaining the unstructured data in storage with an association to the message.16. The non-transitory, machine-readable medium of any one of embodiments 11-15, wherein the one or more actions includes: using the unstructured data to amend one or more dataAttorney Docket No. 201818-0585628 fields of the plurality of data fields where data in the one or more data fields includes structured data, scmi-unstructurcd data, or unstructured data.17. The non-transitory, machine-readable medium of any one of embodiments 11-16, wherein the operations further comprise: obtaining, by the computing system, a plurality of historical messages that were sent via a messaging system and that each include a plurality of data fields, wherein a set of data fields of the plurality of data fields includes unstructured data; obtaining, by the computing system, one or more historical actions taken for a respective historical message of the plurality of historical messages; training, by the computing system and using the one or more historical actions and at least a portion of the unstructured data in at least a portion of the set of data fields of the plurality data fields, the large language model; and saving, by the computing system, the large language model in storage.18. A non-transitory, machine-readable medium storing instructions that, when executed by one or more processors, effectuate operations comprising: obtaining, by a computer system, a plurality of historical messages that were sent via a messaging system and that each include a plurality of data fields, wherein a first set of data fields of the plurality of data fields includes unstructured data; obtaining, by the computer system, one or more historical actions taken for a respective historical message of the plurality of historical messages; training, by the computer system and using the one or more historical actions and at least a portion of the unstructured data in at least a portion of the first set of data fields of the plurality data of fields, a large language model; and saving, by the computer system, the large language model in storage.19. The non-transitory, machine-readable medium of embodiment 18, wherein the operations further comprise: receiving, by the computer system, feedback from decisions produced by the large language model; and updating, by the computer system and based on the feedback, the large language model.20. The non-transitory, machine-readable medium of any one of embodiments 18 or 19, wherein the operations further comprise: compiling the plurality of historical messages by: identifying inbound messages of the plurality of historical messages that include data fieldsAttorney Docket No. 201818-0585628 having unstructured data; identifying outbound messages of the plurality of historical messages that include at least one data field having unstructured data or that do not include a data field that having unstructured data; and linking the inbound messages and the outbound messages based on a reference number included in each inbound message and each outbound message, wherein the training is performed using the linked messages.21. The non-transitory, machine-readable medium of any one of embodiments 18-20, wherein the large language model is trained using a low-rank adaptation technique.

Claims

Attorney Docket No. 201818-0585628CLAIMSWhat is claimed is:

1. A system, comprising: one or more processors; and memory storing instructions that when executed by the one or more processors cause the one or more processors to: receive a message that is destined to one or more recipients and that includes a plurality of data fields, wherein a first set of data fields of the plurality of data fields includes unstructured data; parse the unstructured data in at least a portion of the first set of data fields to obtain parsed data; process, with a large language model, the parsed data to determine an interpretation of the unstructured data and a confidence score in that interpretation; determine, based on the interpretation and the confidence score, one or more actions to take with at least one of the message or the unstructured data in the at least the portion of the first set of data fields; perform the one or more actions; and send the message to the one or more recipients.

2. The system of claim 1, wherein the one or more actions includes: sending to an agent computing device for verification of the interpretation and the one or more actions if the confidence score does not satisfy a threshold.

3. The system of claim 1, wherein the one or more actions includes: providing instructions to automatically perform the one or more actions if the confidence score satisfies a threshold.

4. The system of claim 1, wherein the one or more actions includes:Attorney Docket No. 201818-0585628 retaining the unstructured data as the unstructured data was originally presented in the message.

5. The system of claim 1, wherein the one or more actions includes: stripping the unstructured data from the message; and retaining the unstructured data in storage with an association to the message.

6. The system of claim 1, wherein the one or more actions includes: using the unstructured data to amend one or more data fields of the plurality of data fields where data in the one or more data fields includes structured data, semi-unstructured data, or unstructured data.

7. The system of claim 1, wherein the instructions cause the one or more processors to: obtain a plurality of historical messages that were sent via a messaging system and that each include a plurality of data fields, wherein a set of data fields of the plurality of data fields includes unstructured data; obtain one or more historical actions taken for a respective historical message of the plurality of historical messages; train, using the one or more historical actions and at least a portion of the unstructured data in at least a portion of the set of data fields of the plurality of data fields, the large language model; and save the large language model in storage.

8. The system of claim 7, wherein obtaining the plurality of historical messages includes pairing inbound messages of the plurality of historical messages that include the unstructured data with outbound messages of the plurality of historical messages, and linking the inbound messages and the outbound messages via a reference identifier included in each of the inbound messages and the outbound messages.Attorney Docket No. 201818-05856289. The system of claim 7, wherein training the large language model comprises fine-tuning the large language model using a Low-Rank Adaptation (LoRA) technique.

10. The system of claim 7, wherein the training the large language model further includes generating reasoning using a high-parameter language model and incorporating the reasoning into a training dataset for the large language model.

11. A non-transitory, machine-readable medium storing instructions that, when executed by one or more processors, effectuate operations comprising: receiving, by a computing system, a message that is destined to one or more recipients and that includes a plurality of data fields, wherein a first set of data fields of the plurality of data fields includes unstructured data; parsing, by the computing system, the unstructured data in at least a portion of the first set of data fields to obtain parsed data; processing, by the computing system and with a large language model, the parsed data to determine an interpretation of the unstructured data and a confidence score in that interpretation; determining, by the computing system and based on the interpretation and the confidence score, one or more actions to take with at least one of the message or the unstructured data in the at least the portion of the first set of data fields; performing, by the computing system, the one or more actions; and sending, by the computing system, the message to the one or more recipients.

12. The non-transitory, machine-readable medium of claim 11, wherein the one or more actions includes: sending to an agent computing device for verification of the interpretation and the one or more actions if the confidence score does not satisfy a threshold.

13. The non-transitory, machine-readable medium of claim 11, wherein the one or more actions includes:Attorney Docket No. 201818-0585628 retaining the unstructured data as the unstructured data was originally presented in the message.

14. The non-transitory, machine-readable medium of claim 11, wherein the one or more actions includes: stripping the unstructured data from the message; and retaining the unstructured data in storage with an association to the message.

15. The non-transitory, machine-readable medium of claim 11, wherein the one or more actions includes: using the unstructured data to amend one or more data fields of the plurality of data fields where data in the one or more data fields includes structured data, semi-unstructured data, or unstructured data.

16. The non-transitory, machine-readable medium of claim 11, wherein the operations further comprise: obtaining, by the computing system, a plurality of historical messages that were sent via a messaging system and that each include a plurality of data fields, wherein a set of data fields of the plurality of data fields includes unstructured data; obtaining, by the computing system, one or more historical actions taken for a respective historical message of the plurality of historical messages; training, by the computing system and using the one or more historical actions and at least a portion of the unstructured data in at least a portion of the set of data fields of the plurality data fields, the large language model; and saving, by the computing system, the large language model in storage.

17. A non-transitory, machine-readable medium storing instructions that, when executed by one or more processors, effectuate operations comprising: obtaining, by a computer system, a plurality of historical messages that were sent via a messaging system and that each include a plurality of data fields, wherein a first set of data fields of the plurality of data fields includes unstructured data;Attorney Docket No. 201818-0585628 obtaining, by the computer system, one or more historical actions taken for a respective historical message of the plurality of historical messages; training, by the computer system and using the one or more historical actions and at least a portion of the unstructured data in at least a portion of the first set of data fields of the plurality of data fields, a large language model; and saving, by the computer system, the large language model in storage.

18. The non-transitory, machine-readable medium of claim 17, wherein the operations further comprise: receiving, by the computer system, feedback from decisions produced by the large language model; and updating, by the computer system and based on the feedback, the large language model.

19. The non-transitory, machine-readable medium of claim 17, wherein the operations further comprise: compiling the plurality of historical messages by: identifying inbound messages of the plurality of historical messages that include data fields having unstructured data; identifying outbound messages of the plurality of historical messages that include at least one data field having unstructured data or that do not include a data field that having unstructured data; and linking the inbound messages and the outbound messages based on a reference number included in each inbound message and each outbound message, wherein the training is performed using the linked messages.

20. The non-transitory, machine-readable medium of claim 17, wherein the large language model is trained using a low-rank adaptation technique.

Images