AI native mobile bank intelligent interaction method and system based on AGUI protocol

Abstract

This invention provides an AI-native mobile banking intelligent interaction method and system based on the AGUI protocol. Through an event-driven front-end rendering mechanism, this invention decouples the mobile banking user-side terminal from the bank-side server. The user-side terminal is only responsible for data collection, instruction forwarding, and interface rendering, while the bank-side server uses a large model to understand user intent and generate corresponding cards for AGUI event-driven front-end rendering. It also supports flexible expansion of AGUI event types and streaming rendering, achieving lightweight, highly responsive, and scalable mobile banking interaction, solving the problems of high coupling between the traditional mobile banking front-end and business logic, low iteration efficiency, and long user waiting times.

Description

Technical Field

[0001] This invention relates to the fields of financial technology and human-computer interaction technology, specifically to an AI-native mobile banking intelligent interaction method and system based on the AGUI protocol. Background Technology

[0002] With the application of AI-native technologies in the banking industry, institutions such as ICBC and Shanghai Bank have launched AI-native mobile banking services that support voice commands. However, these services still rely on a customized integration model between traditional GUIs and AI agents. Each type of AI agent (such as wealth management or risk control agents) requires separate development of a front-end interface, resulting in high development costs and low adaptation efficiency. The average development cycle for banks integrating a single type of AI agent is 2-3 weeks, and the interaction logic of different agents is inconsistent, leading to fragmented user experiences (e.g., the interaction process for wealth management consultation differs significantly from the process for fund transfer verification).

[0003] Currently, Shanghai Bank's 2025 AI-native mobile banking only supports "basic business transactions via voice commands." The integration of the AI ​​agent with the GUI requires customized interfaces for each type of business (transfer / wealth management), and adapting one AI agent takes an average of 10 days. Furthermore, there is no state synchronization mechanism (e.g., users need to pull down to refresh to update wealth management returns). ICBC's "ICBC Smart Surge" mobile banking, while integrating a massive model, requires the AI ​​agent's output to be "fully generated and pushed out all at once" (users wait 3-5 seconds), and it does not involve standardized communication using the AGUI protocol. Cross-platform adaptation (iOS / Android / HarmonyOS) requires additional development.

[0004] Meanwhile, there are three major pain points in the interaction between traditional GUIs and intelligent agents:

[0005] Non-standardized communication: There are no unified rules for event transmission and state synchronization between the agent and the GUI. Data formats need to be customized for each agent (e.g., financial agent outputs JSON format, risk control agent outputs XML format), and the front end needs to parse repeatedly.

[0006] Non-real-time interaction: The "one-time request-response" model is adopted. The results generated by the intelligent agent (such as the Q&A content of AI customer service) can only be pushed to the GUI after all the results are completed, resulting in long user waiting time (average 3-5 seconds); the adaptation cost due to differences in system APIs is high and user waiting time is long.

[0007] State desynchronization: The GUI state needs to be manually refreshed (e.g., the user needs to pull down to update the investment income), which cannot be synchronized with the real-time calculated state of the intelligent agent, resulting in "data delay" or "state conflict" (e.g., the user operation conflicts with the investment redemption process executed by the intelligent agent).

[0008] In summary, the existing technology has the following problems: lack of adaptation to financial scenarios, non-real-time interaction between GUI and intelligent agent, non-standardized communication, and asynchronous states. Summary of the Invention

[0009] The purpose of this invention is to address the problems of lack of adaptation to financial scenarios, non-real-time interaction between the GUI and the intelligent agent, non-standardized communication, and asynchronous states.

[0010] To this end, on the one hand, embodiments of the present invention provide an AI-native mobile banking intelligent interaction method based on the AGUI protocol, the method comprising the following steps:

[0011] The terminal collects the raw input data and sends the raw data to the application server;

[0012] The application server converts voice data into text data through a speech recognition module, and then uses a large model intent understanding module to parse the user's business intent and extract key business parameters.

[0013] The application server generates a corresponding AGUI event based on the parsed user business intent and pushes the AGUI event to the terminal.

[0014] The terminal identifies the AGUI event type through an event listener, calls the corresponding rendering function to render and display the interface card;

[0015] After completing the interactive operation on the interface card, the terminal forwards the operation command to the application server, which then distributes it to the AI ​​server.

[0016] The AI ​​server calls the corresponding business intelligence agent to process business logic. During the processing, new AGUI events are generated, and the new AGUI events are forwarded to the application server to drive the terminal to update the interface cards.

[0017] After the business process is completed, the AI ​​server generates a result-type AGUI event through the application server, and the terminal renders the result-type AGUI event to display the result card, thus completing the interaction process.

[0018] On the other hand, embodiments of the present invention provide an AI-native mobile banking intelligent interaction system based on the AGUI protocol, including:

[0019] The acquisition unit is used to acquire raw data input by the terminal and send the raw data to the application server;

[0020] The parsing unit is used by the application server to convert voice data into text data through the voice recognition module, and then to parse the user's business intent through the large model intent understanding module to extract key business parameters.

[0021] The push unit is used by the application server to generate a corresponding AGUI event based on the parsed user business intent, and push the AGUI event to the terminal.

[0022] The rendering unit is used by the terminal to identify the AGUI event type through the event listener, call the corresponding rendering function to render the interface card and display it;

[0023] An interaction unit is used to complete interactive operations on interface cards. The terminal forwards the operation instructions to the application server, which then distributes them to the AI ​​server.

[0024] The processing unit is used by the AI ​​server to call the corresponding business intelligence agent to process business logic, generate new AGUI events during the processing, and forward the new AGUI events to the application server to drive the terminal to update the interface cards.

[0025] The generation unit is used to generate a result-type AGUI event through the application server after the business processing is completed. The terminal then renders the result-type AGUI event to display the result card, thus completing the interaction process.

[0026] In another aspect, embodiments of the present invention provide a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it employs the method described in any one of the above descriptions. The computer program includes an AGUI event parsing algorithm module, a differential state synchronization algorithm module, and a scene weight adjustment algorithm module, and supports incremental rendering of streaming events.

[0027] In another aspect, embodiments of the present invention provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method described in any one of the above descriptions. The electronic device is a smartphone or a smartwatch.

[0028] The above technical solution has the following beneficial effects:

[0029] This invention decouples the user-side terminal from the bank-side server, so that adding new business scenarios does not require modification of the core framework, but only requires extending AGUI events and corresponding cards, improving iteration efficiency by more than 50%.

[0030] This invention significantly optimizes the user interaction experience. Employing streaming push and incremental rendering technologies, it reduces user response time from 3-5 seconds to ≤300 milliseconds. Customer service Q&A is segmented by Chinese characters, and financial information is displayed sequentially by product, completely eliminating long waiting times. It also supports screen reading and recognition of 8 mainstream dialects, increasing the success rate for elderly users from 75%-82% to 95%, thus adapting to a wider user base. The interface and intelligent agent state are precisely synchronized. Through a differential state synchronization mechanism, state consistency reaches 99.99%, completely resolving state conflict issues such as "redemption completed but still showing as redemption in progress" in traditional models. Even in the event of a network interruption, 100% state recovery can be achieved through an event retransmission mechanism.

[0031] This invention features extremely fast response and employs a streaming rendering mechanism, rendering card content in stages, reducing user waiting time to less than 300 milliseconds, resulting in virtually imperceptible waiting.

[0032] AGUI events support unlimited expansion, enabling rapid adaptation to various financial business scenarios such as money transfers, wealth management, loans, and intelligent customer service. User-side terminals do not need to handle complex business logic, reducing client-side development and maintenance costs, while also minimizing resource consumption on the user-side terminal. Attached Figure Description

[0033] Figure 1 This is a flowchart of the AI-native mobile banking intelligent interaction method based on the AGUI protocol provided in this embodiment of the invention;

[0034] Figure 2 This is a schematic diagram of the structure of the AI-native mobile banking intelligent interaction system based on the AGUI protocol provided in an embodiment of the present invention;

[0035] Figure 3 This is a flowchart of the first implementation method of the AI-native mobile banking intelligent interaction method based on the AGUI protocol provided in this invention.

[0036] Figure 4 This is an overall architecture diagram of the AI-native mobile banking intelligent interaction system based on the AGUI protocol provided in this embodiment of the invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] The purpose of this invention is to address the problems of lack of adaptation to financial scenarios, non-real-time interaction between GUI and intelligent agent, non-standardized communication, asynchronous states, and high compliance adaptation costs, and to provide a standardized, real-time, and secure AI-native mobile banking intelligent interaction solution based on the AGUI protocol.

[0039] In embodiments of the present invention, such as Figure 1 This paper provides an AI-native mobile banking intelligent interaction method based on the AGUI protocol, which includes the following steps:

[0040] S101: The terminal collects the input raw data and sends the raw data to the application server;

[0041] S102: The application server converts voice data into text data through a speech recognition module, and then parses the user's business intent through a large-scale model intent understanding module to extract key business parameters; specifically including:

[0042] The converted text data is classified by intent and matched with preset financial business scenario tags, including transfer, wealth management, loan, account inquiry, and intelligent customer service;

[0043] Key business parameters are extracted from text data based on named entity recognition technology. These key business parameters include the transaction object, transaction amount, and product name.

[0044] The extracted key business parameters are matched with bank-side user account data to obtain business-related data such as payee account, user holding information, and loanable amount.

[0045] S103: The application server generates a corresponding AGUI event based on the parsed user business intent and pushes the AGUI event to the terminal;

[0046] The AGUI events,

[0047] This includes at least one of the following: transaction initialization event, risk verification event, service wait event, log audit event, transaction success event, and transaction failure event; different types of AGUI events correspond to different interface cards;

[0048] The aforementioned log audit event is generated upon completion of any complete financial transaction interaction process. This event records key information from the entire transaction process and synchronizes it to the audit system, enabling traceability of business operations. Specifically, based on the event-driven rendering mechanism of the AGUI protocol, AGUI events are categorized according to financial business scenarios. Different events correspond to different front-end interface cards, and unlimited expansion is supported. The following are all AGUI events explicitly mentioned in the document, including event type, triggering scenario, and rendering card details:

[0049] Transfer scenario events

[0050] This scenario includes 6 core event types, covering the UI interaction drivers of the entire transfer process:

[0051] FIN_TRADE_INIT

[0052] Triggering scenario: After the application server parses the user's transfer intent...

[0053] Rendered Card: Transaction confirmation card, displaying the transfer amount, recipient information, and a confirmation button.

[0054] FIN_RISK_CHECK

[0055] Triggering scenario: When the risk control assistant assesses that the transfer carries a medium or higher risk.

[0056] Render Card: Verification card, displaying the verification code input box, countdown timer, and verification prompts.

[0057] FIN_SERVICE_PAUSE

[0058] Triggering scenario: The AI ​​server's transfer assistant begins processing a transfer request and needs to wait for risk control verification.

[0059] Rendering Card: A waiting card that displays a loading animation, a disabled action button, and a "Processing" message.

[0060] FIN_SERVICE_RESUME

[0061] Triggering scenario: When the risk control assistant completes the verification code verification and passes the risk assessment.

[0062] Rendering Card: Processing the card, displaying a transfer progress animation and a "Verification successful, transferring funds" message.

[0063] FIN_TRADE_SUCCESS

[0064] Triggering scenario: When the core host completes account deduction and fund transfer, and the transfer operation is successful.

[0065] Rendering Card: Success Card, displaying a success icon, transfer amount, recipient, transaction number, completion time, and action buttons.

[0066] FIN_TRADE_FAIL

[0067] Triggering scenario: When a core host transfer operation fails (e.g., insufficient balance, incorrect account, etc.).

[0068] Rendering Card: Failure Card, displaying the failure message, reason for failure, and retry button.

[0069] FIN_LOG_AUDIT

[0070] Triggering scenario: When an audit log needs to be recorded after the entire transfer process is completed.

[0071] Function: Generates transaction audit logs and synchronizes them to the bank's audit system; no front-end card rendering is required.

[0072] Financial Management Scenarios and Events

[0073] This scenario includes 5 types of scalable events, supporting the entire process of inquiring about, purchasing, and redeeming wealth management products:

[0074] FIN_PRODUCT_LIST

[0075] Triggering scenario: When a user issues a list search command such as "recommend financial products"

[0076] Rendered Card: A list card for financial products, displaying information such as product name, term, and yield.

[0077] FIN_PRODUCT_DETAIL

[0078] Triggering scenario: When a user clicks on a financial product in the list.

[0079] Rendered Card: Product Details Card, displaying information such as product yield, risk level, investment direction, minimum investment amount, etc.

[0080] FIN_PURCHASE_CONFIRM

[0081] Triggering scenario: When a user confirms the purchase of a certain financial product.

[0082] Rendered Card: Purchase confirmation card, displaying information such as purchase amount, expected returns, and product term.

[0083] FIN_HOLDING_INFO

[0084] Triggering scenario: When a user checks the holdings of their purchased wealth management products.

[0085] Rendered Card: A holdings information card that displays current returns, product shares, and market value of holdings.

[0086] FIN_REDEEM_CONFIRM

[0087] Triggering scenario: When a user applies to redeem their existing holdings in a wealth management product.

[0088] Rendered Card: Redemption confirmation card, displaying information such as redemption amount, estimated arrival time, and transaction fees.

[0089] Loan scenario events

[0090] This scenario includes 6 types of scalable events, covering the entire process of loan inquiry, application, and repayment:

[0091] FIN_LOAN_QUERY

[0092] Triggering scenario: When a user issues a credit limit inquiry command such as "How much can I borrow?"

[0093] Rendered Card: Credit Limit Inquiry Card, displaying information such as available loan amount, loan interest rate, and loan term.

[0094] FIN_LOAN_APPLY

[0095] Triggering scenario: When a user initiates a loan application.

[0096] Rendered Card: Loan application card, displaying information such as amount input box, term selection option, repayment method, etc.

[0097] FIN_CREDIT_CHECK

[0098] Triggering scenario: When the system conducts a credit assessment on a user.

[0099] Rendered Card: Credit Assessment Card, displaying assessment progress, waiting prompts, and estimated completion time.

[0100] FIN_CONTRACT_SIGN

[0101] Triggering scenario: Loan approval is granted, and the user needs to sign an electronic contract.

[0102] Rendered Card: Contract Signing Card, displaying the contract content, electronic signature area, and signing instructions.

[0103] FIN_REPAY_REMIND

[0104] Triggering scenario: When the system sends a reminder near the loan repayment date.

[0105] Rendered Card: A repayment reminder card that displays information such as the amount due, repayment date, and repayment account.

[0106] FIN_REPAY_SUCCESS

[0107] Triggering scenario: When a user completes a loan repayment and the repayment is successful.

[0108] Rendered Card: A card indicating successful repayment, displaying information such as the amount repaid, remaining principal, and repayment date.

[0109] Account query scenario event

[0110] This scenario includes four types of scalable events, supporting the query of account information and transaction records:

[0111] FIN_BALANCE_SHOW

[0112] Triggering scenario: When a user issues an account balance inquiry command such as "check balance".

[0113] Rendering Card: Balance Display Card, showing the balance information of each account under the username.

[0114] FIN_TRANSACTION_LIST

[0115] Triggering scenario: When a user issues a transaction record query command such as "view transaction history".

[0116] Rendered Card: Transaction history card, displaying income and expenditure list, transaction time, and filtering functions.

[0117] FIN_TRANSACTION_DETAIL

[0118] Triggering scenario: When a user clicks on a transaction record to view details...

[0119] Rendered Card: Transaction details card, displaying information such as transaction time, amount, counterparty account, and transaction type;

[0120] FIN_ACCOUNT_OVERVIEW

[0121] Triggering scenario: When a user issues an asset summary query command such as "My Assets".

[0122] Rendered Card: Asset Overview Card, displaying summary information on assets and liabilities such as deposits, wealth management products, and liabilities.

[0123] Intelligent customer service scenario events

[0124] This scenario includes four types of scalable events to support intelligent consultation and human service transfer:

[0125] FIN_FAQ_ANSWER

[0126] Triggering scenario: When users ask common financial business questions;

[0127] Rendered cards: Question and answer cards that display the question, answer, and relevant business links;

[0128] FIN_GUIDE_STEPS

[0129] Triggering scenario: When users need guidance on business operations;

[0130] Rendering Card: Operation guide card, displaying a list of steps, illustrations, and precautions;

[0131] FIN_HUMAN_TRANSFER

[0132] Triggering scenario: When the intelligent customer service cannot answer a user's question and needs to be transferred to a human agent;

[0133] Rendered cards: Convert to human-generated cards, displaying the waiting queue position, estimated waiting time, and queuing reminders;

[0134] FIN_FEEDBACK_FORM

[0135] Triggering scenario: When a user needs to file a complaint or submit a suggestion;

[0136] Rendering Card: Feedback form card, displaying the question type selection, content input box, and contact information field;

[0137] Different types of AGUI events correspond to different interface cards.

[0138] S104: The terminal identifies the AGUI event type through the event listener, calls the corresponding rendering function to render the interface card and display it;

[0139] S105: After completing the interactive operation on the interface card, the terminal forwards the operation instruction to the application server, which then distributes it to the AI ​​server.

[0140] S106: The AI ​​server calls the corresponding business intelligence agent to process business logic. During the processing, a new AGUI event is generated, and the new AGUI event is forwarded to the application server to drive the terminal to update the interface card.

[0141] S107: After the business process is completed, the AI ​​server generates a result-type AGUI event through the application server, and the terminal renders the result-type AGUI event to display the result card, thus completing the interaction process.

[0142] Among them, such as Figure 4The application server, as the core relay node of the bank's server-side, is a key hub connecting user-side terminals, AI servers, and core hosts. Its complete module composition and functional description are as follows:

[0143] 1) Encrypted communication and protocol adaptation module

[0144] This module provides the basic communication support for the application server, corresponding to the encryption / decryption (SM4) and protocol adaptation functions marked in the architecture diagram.

[0145] Responsible for establishing HTTPS / WebSocket / TLS encrypted communication links with user-side terminals, and performing SM4 algorithm encryption and decryption on transmitted user input data and AGUI event data to ensure data transmission security.

[0146] It completes the adaptation and conversion of multiple protocols, supports the conversion of user-side request data into a format that can be recognized by the bank's internal system, and converts internal AGUI event data into a format that can be parsed by the front end, so as to achieve smooth communication across nodes.

[0147] 2) Automatic Speech Recognition (ASR) Module

[0148] The corresponding speech-to-text (ASR) function marked in the architecture diagram is specifically designed to handle users' voice input.

[0149] It receives raw voice data forwarded from user-side terminals and converts colloquial voice commands into standardized text commands using speech recognition technology.

[0150] The output text commands are directly transmitted to the large model intent understanding module, providing basic data for subsequent intent parsing and adapting to the voice interaction scenario of mobile banking.

[0151] 3) Large Model Intent Understanding Module

[0152] The large model's intent-understanding function, as marked in the architecture diagram, is the core module for realizing AI-native intelligent interaction.

[0153] It receives text commands output by the speech recognition module, or text input directly forwarded by the user, and performs three core operations:

[0154] Business scenario matching: Categorize user commands into preset financial scenarios, such as money transfer, wealth management recommendations, loan inquiries, etc.

[0155] Key parameter extraction: Extract core business information from instructions, such as the amount and recipient in transfer scenarios, and the risk level and product term in wealth management scenarios.

[0156] User data association: Retrieve user account association data, such as payee account and user risk assessment level, to form a complete business analysis result.

[0157] 4) AGUI Event Generation and Routing Module

[0158] The AGUI event generation and event routing functions, as marked in the architecture diagram, are key modules that connect business parsing and front-end rendering.

[0159] Event generation: Based on the parsing results of the large model intent understanding module, the business processing feedback of the AI ​​server, and the operation results of the core host, standardized AGUI events are generated, such as transaction initialization events, risk verification events, and transaction success events, and unique identifiers and business parameters are configured for each event.

[0160] Event routing: Based on event type and business requirements, AGUI events are accurately pushed to the corresponding nodes, including pushing them to the user-side terminal to drive interface rendering, or synchronizing them to the AI ​​server and core host to trigger subsequent business processes.

[0161] 5) Streaming data processing module

[0162] This module supports the core mechanism of streaming rendering in documents and is a key support for achieving ultra-fast front-end response.

[0163] Receive complete event data output from the AGUI event generation module, and according to the rendering logic of the front-end card, split the data into multiple ordered fragments, such as amount information fragments, payee information fragments, and button state fragments.

[0164] The segmented data is transmitted to the encrypted communication module in a preset order and pushed to the user terminal step by step via the WebSocket protocol. At the same time, a "transmission complete" mark is added to the last data segment to ensure that the front end can determine the integrity of the card rendering.

[0165] Module collaboration workflow:

[0166] The encrypted communication module receives voice / text data from the user side, decrypts it, and then transmits it to the voice recognition module (voice data) or directly to the large model intent understanding module (text data).

[0167] After the large model intent understanding module completes the parsing, it passes the results to the AGUI event generation module.

[0168] The AGUI event generation module generates corresponding events, and if streaming rendering is required, the data is split by the streaming data processing module.

[0169] The encrypted communication module encrypts AGUI event data and pushes it to the user terminal, while simultaneously synchronizing it to relevant backend nodes through the event routing module.

[0170] In embodiments of the present invention, such as Figure 2 It also provides an AI-native mobile banking intelligent interaction system based on the AGUI protocol, including:

[0171] The acquisition unit 21 is used to acquire raw data input by the terminal and send the raw data to the application server;

[0172] The parsing unit 22 is used by the application server to convert voice data into text data through the voice recognition module, and then to parse the user's business intent through the large model intent understanding module to extract key business parameters.

[0173] The push unit 23 is used by the application server to generate a corresponding AGUI event based on the parsed user business intent, and push the AGUI event to the terminal.

[0174] Rendering unit 24 is used by the terminal to identify the AGUI event type through the event listener, call the corresponding rendering function to render the interface card and display it;

[0175] Interaction unit 25 is used to complete interactive operations on the interface card. The terminal forwards the operation instructions to the application server, which then distributes them to the AI ​​server.

[0176] Processing unit 26 is used by the AI ​​server to call the corresponding business intelligence agent to process business logic, generate new AGUI events during the processing, and forward the new AGUI events to the application server to drive the terminal to update the interface cards;

[0177] The generation unit 27 is used to generate a result-type AGUI event through the application server after the business processing is completed, and the terminal renders the result-type AGUI event to display the result card, thus completing the interaction process.

[0178] The AI-native mobile banking intelligent interaction system based on the AGUI protocol adopts the aforementioned AI-native mobile banking intelligent interaction method based on the AGUI protocol. Its principle and process are the same as those of the AI-native mobile banking intelligent interaction method based on the AGUI protocol, and will not be repeated here.

[0179] In this embodiment of the invention, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it employs the method described in any one of the above descriptions. The computer program includes an AGUI event parsing algorithm module, a differential state synchronization algorithm module, and a scene weight adjustment algorithm module, and supports incremental rendering of streaming events.

[0180] In this embodiment of the invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method described in any of the above-mentioned embodiments. The electronic device is a smartphone or a smartwatch.

[0181] Smartphone front-facing camera with a resolution of ≥1080P;

[0182] The smartwatch supports HarmonyOS Connect, enabling cross-device synchronization of AGUI events with smartphones.

[0183] Example 1:

[0184] This invention provides an AI-native mobile banking intelligent interaction method based on the AGUI protocol, the implementation steps of which include:

[0185] (I) System Deployment

[0186] 1. User-side terminal: Deploy a mobile banking APP with built-in event listeners and card rendering modules, and pre-register the AGUI event-card mapping relationship for transfer scenarios, including transaction confirmation card, verification card, waiting card, success card, etc.

[0187] 2. Bank-side server

[0188] Application server: Deploy the Automatic Speech Recognition (ASR) module and the large model intent understanding module, and configure the WebSocket communication protocol;

[0189] AI Server: Deploys intelligent agents for fund transfer and risk control, responsible for process handling and risk assessment of fund transfer transactions;

[0190] Core host: Deploys the accounting system and is responsible for account deductions and accounting operations.

[0191] (II) Interaction Process

[0192] Input collection and forwarding: When a user opens the mobile banking app, for example, by voice inputting "Transfer 5,000 yuan to Li Si for me", the app collects the voice data and forwards it to the bank's application server.

[0193] Intent understanding and event generation: The application server converts speech into text through the ASR module, and then uses a large model to parse the user's intent as "transfer 5,000 yuan to Li Si", and extracts key parameters such as the recipient's account number; then it generates a FIN_TRADE_INIT event, carrying data such as amount and recipient information.

[0194] Streaming rendering of confirmation cards: The application server splits the dataset of the FIN_TRADE_INIT event into three segments: amount, payee, and confirmation button, and pushes them to the user's app in segments; the app renders the amount area after receiving the first segment, the payee area after receiving the second segment, and the confirmation button after receiving the third segment. The entire rendering process takes about 0.3 seconds.

[0195] User interaction and command forwarding: When a user clicks the confirmation button, the APP forwards the confirmation command to the bank's AI server.

[0196] Business processing and risk verification: After receiving the instruction, the AI ​​server's transfer agent first checks the user's account balance and the validity of the payee's account, and then generates a FIN_SERVICE_PAUSE event to drive the APP to render the waiting card;

[0197] The transfer agent requests the risk control agent to conduct a risk assessment. The risk control agent gives a risk score based on factors such as the user's login device and transfer habits, generates a FIN_RISK_CHECK event, and drives the APP to render the SMS verification card.

[0198] Verification passed and core operations: The user enters the SMS verification code and submits it. After the risk control agent verifies it, it generates a FIN_SERVICE_RESUME event, which drives the APP to render and process the card. The transfer agent requests the core host to execute the transfer operation. The core host deducts 5,000 yuan from the user's account, adds 5,000 yuan to Li Si's account, and generates a transaction serial number.

[0199] Feedback: After the transfer is completed, the transfer agent generates a FIN_TRADE_SUCCESS event through the application server, carrying data such as amount, recipient, transaction number, and time. After receiving the event, the APP streams and renders the transfer success card, displays the electronic receipt, and the user can click "Complete" or "Share Receipt" to end the entire interaction process.

[0200] Log auditing: The application server generates a FIN_LOG_AUDIT event to record the complete information of this transfer to the auditing system.

[0201] (III) Business Scenario Expansion

[0202] This invention can be quickly extended to business scenarios such as wealth management and lending. For example, in a wealth management scenario, a new FIN_PRODUCT_LIST event and a list of wealth management products are added. When a user inputs "recommend wealth management products" via voice, the application server understands the intent through a large model and generates the corresponding event, driving the APP to render the list of wealth management products, thus realizing intelligent interaction in wealth management business.

[0203] The AGUI protocol of this invention features flexible and scalable event types. Adding interactive capabilities to any financial business scenario requires only three standardized steps without altering the core framework of the user-side terminal and the bank-side server: First, the AGUI event management and generation module on the bank-side server defines a unique event type name and corresponding data structure based on the functional requirements of the new business scenario, clarifying the event triggering conditions and required business parameters, providing accurate data for front-end interface rendering. Second, the user-side terminal develops an interface card component matching the new event type in its local card component library, specifically designing the card's display area, interactive elements, and visual rendering style to ensure a high degree of alignment between card functionality and the actual needs of the business scenario. Third, the user-side terminal's event listener completes the association registration between the new event type and the corresponding card rendering function, establishing a one-to-one mapping relationship. This allows the terminal to automatically call the specified rendering function to load and display the interface card when it receives this type of AGUI event, thus achieving non-intrusive addition of business scenarios and enabling rapid system iteration and flexible expansion.

[0204] Example 2:

[0205] This invention provides an AI-native mobile banking intelligent interaction system based on the AGUI protocol, such as... Figure 3 ,include:

[0206] Step S1: User Input Collection and Forwarding

[0207] When a user opens the mobile banking app, triggers the voice input function, and says, "Recommend low-risk short-term wealth management products," the app's voice acquisition module collects the voice input data in real time and forwards the raw voice data directly to the bank's application server via an encrypted network, without any local intent parsing or business processing.

[0208] Step S2: Speech-to-Text and Intent Analysis

[0209] The application server uses a speech recognition module to convert the received raw speech data into standardized text commands.

[0210] The large model intent understanding module parses the text commands, first matching them to the business scenario of financial recommendation, then extracting the screening conditions of "low risk" and "short term" financial products from the commands, and finally retrieving the user's account association data to obtain the user's risk assessment level as conservative, confirming that the user's risk level is suitable for low-risk financial products.

[0211] Step S3: Generate AGUI events and stream them.

[0212] Based on the intent parsing results, the application server generates an AGUI event for a list of financial products and splits the complete data corresponding to the event into three fragments: product name and product term information, product yield and risk level information, and product operation button information.

[0213] The server pushes three data segments sequentially to the user's terminal via streaming: the product name and term information at 0 milliseconds, the product yield and risk level information at 50 milliseconds, and the product operation button information at 100 milliseconds. At the same time, a data transmission completion indicator is added to each segment.

[0214] Step S4: Front-end event listening and card rendering

[0215] The app's event listener continuously monitors the communication connection with the server. After receiving segmented data pushed in real time, it executes the following sub-steps in sequence:

[0216] Receive and parse the data in each segment, separate the AGUI events with the event type of financial product list, and the financial business-related information corresponding to each segment;

[0217] Based on the parsed event type, query the correspondence between the locally stored AGUI events and rendering resources, and match the financial product list card and dedicated rendering logic corresponding to the event;

[0218] Streaming rendering is performed according to the order of data push. The product name and term area are rendered at 0 milliseconds, the yield and risk level area are rendered at 50 milliseconds, and the product operation button area is rendered at 100 milliseconds. At the same time, a fade-in transition animation is added to improve the interface display effect.

[0219] After detecting the transmission completion marker in the data segment, it is determined that all data of the card has been received, and then the click interaction function of the operation button on the card is activated;

[0220] The rendered list of financial products is then mounted onto the current display screen of the app, completing the final display of the cards.

[0221] Step S5: User Interaction and Command Forwarding

[0222] When a user selects a target financial product from the displayed product list cards and clicks the corresponding "View Details" button, the app will forward information including the unique identifier of the financial product and the user's intention to view the details to the bank's AI server via an encrypted network.

[0223] Step S6: Business Intelligent Agent Processing and Event Generation

[0224] After receiving the user's instruction to view details, the financial management intelligent agent immediately retrieves the complete information of the financial product, generates a product details AGUI event and pushes it to the user's terminal, driving the APP to render the product details card and display all information such as the minimum purchase amount, expected rate of return, investment direction, and product duration to the user;

[0225] If a user clicks the "Buy Now" button in the product details card, the APP forwards the user's purchase instruction to the AI ​​server. The financial management AI then initiates a risk matching verification request to the risk control AI. The risk control AI combines the user's risk level and the product's risk level to complete the verification. After determining that the two match, it generates a service waiting AGUI event, which drives the APP to render the "Processing" waiting card.

[0226] After completing the risk verification, the financial management agent initiates a formal request to the core host to purchase financial products. Upon receiving the request, the core host executes the core operations of deducting funds from the user's account and registering the financial product shares.

[0227] Step S7: Results Feedback and Log Auditing

[0228] After the core host completes the fund deduction and share registration, it returns the purchase success processing result to the wealth management intelligent agent. Based on this, the wealth management intelligent agent generates a transaction success AGUI event, which carries key information such as product name, actual purchase amount, registered shares, and purchase confirmation date. The server sends the event data to the user terminal through streaming push, driving the APP to render the wealth management purchase success card.

[0229] Simultaneously, the application server generates a log auditing AGUI event, fully recording the entire process of this wealth management purchase, from user input to transaction completion, into the bank's auditing system. This completes the entire intelligent interactive process of the wealth management purchase. This embodiment verifies the non-intrusive scenario expansion characteristics of the present invention through a wealth management purchase business scenario. When adding new wealth management services, there is no need to modify the core framework of the mobile banking client and the bank-side server; only the definition of the new AGUI event, the development of the corresponding card component, and the association registration of the event and rendering resources are required.

[0230] The streaming rendering mechanism keeps the overall rendering time of the wealth management product list cards within 120 milliseconds, eliminating the user's sense of waiting and significantly improving interface responsiveness. Based on the AGUI protocol, the event-driven interaction mode realizes a one-stop intelligent service of "voice command initiation → intelligent product recommendation → product details viewing → one-click purchase confirmation". Compared with the traditional mobile banking wealth management purchase process that requires multiple page jumps, manual filtering, and step-by-step filling, the operation steps are reduced by 60%, effectively simplifying user operations and significantly improving the interactive experience of mobile banking wealth management business.

[0231] This invention provides a multimodal push and scenario-based weight adjustment algorithm for AGUI streaming events, covering push logic and weight rules. The AGUI protocol enables a real-time risk control closed loop of "risk detection - user verification - business execution - log auditing", reducing the verification process time from 10 seconds to 2 seconds.

[0232] This invention enables real-time collaboration between risk control and users through FIN_RISK_CHECK and FIN_SERVICE_RESUME events, and links log events with blockchain auditing; this invention's AGUI event-driven risk control closed-loop process, along with the association logic between risk scoring and event push, covers process steps and event interactions.

[0233] Adaptation efficiency is improved exponentially:

[0234] The integration cycle between intelligent agents and mobile banking GUI has been shortened from the traditional 14 days (Shanghai Bank model) to 1 day, reducing adaptation costs by 90%. It supports "zero customization" access across intelligent agents (LangGraph / CrewAI / DeepSeek) and platforms. In a real test, a bank only took 5 days to integrate 6 types of intelligent agents (transfer / wealth management / risk control / customer service / loan / deposit), saving 85% of the time compared to the traditional model.

[0235] The user experience has been significantly improved.

[0236] The response results of the intelligent agent are pushed through streaming AGUI events, reducing the user waiting time from 3-5 seconds (ICBC's "ICBC Smart Surge" mode) to ≤300 milliseconds. The real-time performance of AI customer service Q&A scores 4.8 / 5 (user survey). The success rate of operation for elderly users (over 60 years old) has increased to 95%, which is 20 percentage points higher than that of traditional mobile banking (75%), mainly due to the adaptation of screen reading.

[0237] State consistency and accuracy assurance:

[0238] By adopting a differential state synchronization mechanism, the state consistency between the GUI and the agent reaches 99.99%, and no "state conflict" or "data delay" issues occur. In a real test of 100,000 transactions by a bank, only one transaction caused state inconsistency due to network interruption, with a recovery rate of 100% (through the AGUI event retransmission mechanism).

[0239] Double the efficiency of risk control closed loop:

[0240] The risk control verification process for abnormal transactions has been shortened from an average of 10 seconds (traditional mode) to 2 seconds, while the accuracy rate for identifying fraudulent transactions remains at 97.3% (consistent with the independent test results of the GAT-LSTM risk control model); risk control logs are traced throughout the entire process via the FIN_LOG_AUDIT event, improving audit efficiency by 60% (without the need to manually associate core system logs).

[0241] Comprehensive coverage of financial scenarios:

[0242] The financial-specific AGUI events cover more than 95% of the core businesses of mobile banking (transfer / wealth management / risk control / customer service / loan / deposit).

[0243] Comparison of core metrics with traditional solutions:

[0244]

[0245] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-native mobile banking intelligent interaction method based on the AGUI protocol, characterized in that, The method includes the following steps: The terminal collects the raw input data and sends the raw data to the application server; The application server converts voice data into text data through a speech recognition module, and then uses a large model intent understanding module to parse the user's business intent and extract key business parameters. The application server generates a corresponding AGUI event based on the parsed user business intent and pushes the AGUI event to the terminal. The terminal identifies the AGUI event type through an event listener, calls the corresponding rendering function to render and display the interface card; After completing the interactive operation on the interface card, the terminal forwards the operation command to the application server, which then distributes it to the AI ​​server. The AI ​​server calls the corresponding business intelligence agent to process business logic. During the processing, new AGUI events are generated, and the new AGUI events are forwarded to the application server to drive the terminal to update the interface cards. After the business process is completed, the AI ​​server generates a result-type AGUI event through the application server, and the terminal renders the result-type AGUI event to display the result card, thus completing the interaction process.

2. The AI-native mobile banking intelligent interaction method based on the AGUI protocol according to claim 1, characterized in that, The AGUI events include at least one of the following: transaction initialization event, risk verification event, service wait event, log audit event, transaction success event, and transaction failure event; Different types of AGUI events correspond to different interface cards; The log audit event is generated after the completion of any financial business interaction process, records key information of the entire business process and synchronizes it to the audit system to achieve business operation traceability.

3. The AI-native mobile banking intelligent interaction method based on the AGUI protocol according to claim 1, characterized in that, The application server converts speech data into text data through a speech recognition module, and then uses a large-scale model intent understanding module to parse the user's business intent and extract key business parameters, including: The converted text data is classified by intent and matched with preset financial business scenario tags, including transfer, wealth management, loan, account inquiry, and intelligent customer service; Key business parameters are extracted from text data based on named entity recognition technology. These key business parameters include the transaction object, transaction amount, and product name. The extracted key business parameters are correlated and matched with bank-side user account data to obtain business-related data such as payee account, user holding information, and loanable amount.

4. An AI-native mobile banking intelligent interaction system based on the AGUI protocol, characterized in that: include: The acquisition unit is used to acquire raw data input by the terminal and send the raw data to the application server; The parsing unit is used by the application server to convert voice data into text data through the voice recognition module, and then to parse the user's business intent through the large model intent understanding module to extract key business parameters. The push unit is used by the application server to generate a corresponding AGUI event based on the parsed user business intent, and push the AGUI event to the terminal. The rendering unit is used by the terminal to identify the AGUI event type through the event listener, call the corresponding rendering function to render the interface card and display it; An interaction unit is used to complete interactive operations on interface cards. The terminal forwards the operation instructions to the application server, which then distributes them to the AI ​​server. The processing unit is used by the AI ​​server to call the corresponding business intelligence agent to process business logic, generate new AGUI events during the processing, and forward the new AGUI events to the application server to drive the terminal to update the interface cards. The generation unit is used to generate a result-type AGUI event through the application server after the business processing is completed. The terminal then renders the result-type AGUI event to display the result card, thus completing the interaction process.

5. The AI-native mobile banking intelligent interaction system based on the AGUI protocol according to claim 4, characterized in that, The AGUI events include at least one of the following: transaction initialization event, risk verification event, service wait event, log audit event, transaction success event, and transaction failure event; Different types of AGUI events correspond to different interface cards; The log audit event is generated after the completion of any financial business interaction process, records key information of the entire business process and synchronizes it to the audit system to achieve business operation traceability.

6. The AI-native mobile banking intelligent interaction system based on the AGUI protocol according to claim 4, characterized in that, The parsing unit includes: This is used to classify the converted text data by intent and match it with preset financial business scenario tags, including transfer, wealth management, loan, account inquiry, and intelligent customer service; This is used to extract key business parameters from text data based on named entity recognition technology. The key business parameters include the transaction object, transaction amount, and product name. This is used to associate and match the extracted key business parameters with bank-side user account data to obtain payee account information, user holding information, and loanable amount business-related data.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it employs the method described in any one of claims 1 to 3. The computer program includes an AGUI event parsing algorithm module, a differential state synchronization algorithm module, and a scene weight adjustment algorithm module, and supports incremental rendering of streaming events.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 3, and the electronic device is a smartphone or a smartwatch.

Images