Abstract: A method may include receiving a recommendation request from a computing device, the recommendation request formatted as an application programming interface (API) call with an account identifier, an intended recipient identifier, and an intended action; selecting a machine learning model from a plurality of machine learning models based on the account identifier, intended recipient identifier and intended action; generating an input data structure based on the recommendation request according to an input format of the machine learning model; executing the machine learning model with the input data structure; subsequent to the executing, modifying the intended action to a recommended action based on an output of the machine learning model; transmitting the recommended action as a response to the API call.

Abstract: Systems and techniques to implement a multi-model agent are described herein. When a request corresponding to a session and having a context is received from a user, a large language model (LLM) can be invoked on the request and the context to produce a sequence of actions to complete the request. A first action from a set of available actions cab be selected as a next action along with a first user device from a plurality of user devices to complete the next action. If an indication that the first user device is not available is received, a second action can be selected for the next action on the indication, the request, or the context. A second device can be selected based on the second action and an interface on the second device can be invoked to perform the second action as the next action.

Abstract: A method for facilitating predictive caching of data is provided. The method includes retrieving raw data relating to user activity for a plurality of users, the user activity including a history of web resources accessed by a user; converting the raw data into a structured data set based on a predetermined criterion; generating a model based on the structured data set; training the model by using a training data set, the training data set including the user activity for a predetermined period of time; determining, by using the trained model, a predicted first web resource for the user; and automatically caching, in a memory, the predicted first web resource.

Abstract: A method for facilitating predictive caching of data is provided. The method includes retrieving raw data relating to user activity for a plurality of users, the user activity including a history of web resources accessed by a user; converting the raw data into a structured data set based on a predetermined criterion; generating a model based on the structured data set; training the model by using a training data set, the training data set including the user activity for a predetermined period of time; determining, by using the trained model, a predicted first web resource for the user; and automatically caching, in a memory, the predicted first web resource.

Abstract: Embodiments of present invention are directed to document authentication based on video captures. Instructions may be generated to guide a remotely situated customer through an authentication process in which the customer presents a document (e.g., a driver license, passport, or payment card) to a video capture element of a personal computing device (e.g., a smart phone or tablet computer) in a specified manner so that identifiable feature(s) of that document can be captured in a video file. The video file may then be uploaded to a central server for further processing or archiving. The instructions that guide the customer through the document authentication steps may be either generated locally on the personal computing device via a mobile application or delivered from a remote computer via an Internet browser. With this technique, a banking customer need not physically visit a bank to authenticate certain documents.