In the fields of data weaving (unified management and intelligent scheduling of global data) and edge computing (processing data close to the source), large model services MAAS (standardized intelligent capability output) and large model Agent (autonomous decision-making execution) form a “capability support-task closed loop” collaborative relationship. In the telecommunications industry, the combination of the two addresses business characteristics (multiple heterogeneous data sources and high real-time network requirements) to solve pain points in traditional processes such as “low efficiency of data interconnection and delayed edge decision-making.” Specific applications include:
1. The Role of MAAS and Large Model Agent in Data Weaving and Telecommunications Scenarios
The core aspects of data weaving include data discovery and association, dynamic integration, and unified access. MAAS provides foundational intelligent capabilities such as “semantic understanding and rule generation,” while the large model Agent achieves “full-process automated scheduling and adaptive optimization.”
1.1 Global Data Intelligent Discovery and Lineage Association
MAAS Role: By utilizing natural language understanding and metadata parsing models, it automatically identifies data assets from the dispersed data sources in telecommunications (core network databases, edge base station logs, user APP tracking), extracting metadata such as field meanings and storage locations, and constructs cross-system data lineage relationships based on semantic associations (e.g., “‘User ID’ in CRM is ‘cust_id’, in the billing system it is ‘user_no’”).
Large Model Agent Role: It autonomously schedules data scanning tools (Apache Atlas) to batch collect full-link metadata. Based on the lineage association rules generated by MAAS, it visualizes the construction of a global data map. When the structure of a data source changes (e.g., the base station log adds a “5G signal frequency band” field), it automatically updates the lineage relationship and pushes impact alerts (e.g., “This field change will affect the generation of network optimization reports”).
Telecommunications Business Scenario:
Scenario Description: A provincial telecommunications company needs to integrate three types of heterogeneous data: “core network user data, edge base station operational data, and customer complaint data,” clarifying the relationships between the data (e.g., “Does weak signal from a certain base station lead to an increase in user complaints?”) to support cross-departmental data analysis.
Model Used: MAAS calls the Alibaba Cloud Tongyi Qianwen metadata parsing API (to identify cross-system field associations), and the large model Agent constructs a “data weaving map Agent” based on Huawei Cloud’s Pangu industry large model.
Role: MAAS identifies semantic associations between the core network’s “user belonging base station ID” field and the edge base station’s “base station number” field, determining them to be the same entity; it extracts the keyword “poor signal” from customer complaint texts and associates it with the base station’s “signal strength” field;
The Agent automatically scans metadata from over 100 core network nodes and over 5000 edge base stations, completing a lineage map containing over 2000 data nodes within a week. The accuracy of data association improved from the traditional manual rate of 72% to 96%. The network department quickly identified the causal relationship of “weak signal from a certain area base station → increase in user complaints” through the map, optimizing efficiency by three times.
1.2 Dynamic Data Integration and Format Adaptation
MAAS Role: It calls data transformation models to automatically generate data transformation rules (field mapping, format validation logic) based on telecommunications business scenario requirements (e.g., “Convert the JSON format of edge base station logs to MySQL format for the core network database”), and supports the structured processing of unstructured data (customer call recordings) by transcribing it into text and extracting keywords.
Large Model Agent Role: Based on the transformation rules generated by MAAS, it autonomously schedules integration tools (Flink CDC) to execute cross-system data synchronization, monitoring the status of integration tasks in real-time (e.g., “Is the synchronization delay from edge to core network exceeding 10 seconds?”). When format incompatibility occurs (field type mismatch), it automatically calls MAAS to regenerate adaptation rules, ensuring continuity of data integration.
Telecommunications Business Scenario:
Scenario Description: The telecommunications company needs to synchronize the “5G user access logs” (JSON format, generating 100,000 entries per second) collected in real-time from edge base stations to the core network data warehouse (MySQL format) for real-time analysis of user access success rates, with synchronization delays controlled within 5 seconds.
Model Used: MAAS calls the Baidu Wenxin Yiyan data transformation model (to generate cross-format mapping rules), and the large model Agent constructs a “dynamic integration Agent” based on iFLYTEK’s Spark real-time scheduling model.
Role: MAAS analyzes the differences between the edge log’s “user_access_time” (timestamp format) and the core network’s “access_datetime” (datetime format), generating transformation rules: access_datetime=FROM_UNIXTIME(user_access_time/1000), while validating the value range of the “signal strength” field (0-100);
After deploying the integration task, the Agent monitors synchronization delays in real-time. When a new field “band_5g” (5G frequency band) is added to the base station log, it automatically calls MAAS to generate the MySQL mapping rule for that field (VARCHAR(20)), keeping synchronization delays stable at 3 seconds, improving the real-time analysis of user access success rates by 80%.
1.3 Unified Data Access and Permission Control
MAAS Role: It converts the natural language query requirements of business personnel (e.g., “Query the number of 5G user access failures in a certain area in the last hour”) into standardized query statements (cross-data source SQL), and generates desensitization rules based on data sensitivity levels (e.g., “User phone number is highly sensitive”).
Large Model Agent Role: It constructs a unified data access entry, verifies user permissions (e.g., “The network department can only query base station data and cannot access sensitive user information”), executes the query statements generated by MAAS, and returns desensitized results while recording access logs (e.g., “Who queried what data at what time”), supporting compliance audits.
Telecommunications Business Scenario:
Scenario Description: The telecommunications marketing department needs to query the “distribution of reasons for 5G user access failures in a certain city over the past 3 days,” requiring cross-referencing of edge base station logs (types of access failures) and core network user data (types of user packages), while ensuring that sensitive information such as user phone numbers is not accessible.
Model Used: MAAS calls the Tencent Mix Yuan natural language to SQL model (to generate cross-source query statements), and the large model Agent constructs a “unified access Agent” based on Zhizhu AI’s permission control model.
Role: MAAS transforms the marketing department’s requirements into cross-source SQL: SELECT access_failure_type, package_type, COUNT(*) FROM edge_base_station_logs JOIN core_network_user_data ON base_station_ID = belonging_base_station_ID WHERE time >= DATE_SUB(NOW(), INTERVAL 3 DAY) GROUP BY access_failure_type, package_type, while generating phone number desensitization rules;
After verifying the marketing department’s permissions, the Agent executes the query, automatically desensitizing user phone numbers in the results, reducing query response time from the traditional 2 minutes for cross-system queries to 15 seconds, meeting business needs while ensuring data security.
2. The Role of MAAS and Large Model Agent in Edge Computing and Telecommunications Scenarios
The core aspects of edge computing include real-time processing of edge data, edge-cloud collaboration, and optimization of edge resources. MAAS provides capabilities for “real-time inference and model compression,” while the large model Agent achieves “autonomous decision-making and dynamic scheduling.”
2.1 Real-time Processing of Edge Data and Anomaly Detection
MAAS Role: It provides lightweight inference models (compressed to fit edge computing power) to perform real-time analysis of telecommunications edge data (base station signal data, user access logs), identifying abnormal features (e.g., “base station signal strength drops by 50%” or “user access failure rate exceeds 20%”), and generating anomaly determination rules.
Large Model Agent Role: It deploys MAAS’s lightweight models to edge nodes (base station edge servers), receiving edge data in real-time and executing inference. When anomalies are detected, it autonomously triggers local responses (e.g., “temporarily switch users to neighboring base stations”), while synchronizing anomaly information to the cloud to avoid delays caused by relying on cloud decisions.
Telecommunications Business Scenario:
Scenario Description: The telecommunications company needs to detect “5G user access anomalies” (access failures, frequent disconnections) in real-time at edge base stations. When the anomaly rate exceeds 10%, immediate action is required to avoid user complaints, with a processing delay of <100ms (cloud round-trip delay is about 500ms, which cannot meet the requirement).
Model Used: MAAS provides Huawei Cloud’s Pangu lightweight edge inference model (compressed to <100MB, suitable for edge computing power), and the large model Agent constructs an “edge anomaly handling Agent” based on SenseTime’s edge decision model.
Role: The MAAS model analyzes user access logs in real-time at the edge, identifying “access request timeouts” and “signal strength <30dBm” as abnormal features, setting a response trigger for when the anomaly rate exceeds 10%;
After deploying the Agent to over 5000 edge base stations, a certain area base station detected an access failure rate of 15%, immediately triggering local policy: temporarily switching users in that area to neighboring base stations within 2 kilometers, with a processing delay of only 60ms, resulting in a 75% decrease in user complaints without waiting for cloud instructions.
2.2 Edge-Cloud Collaboration and Model Iteration
MAAS Role: It trains a general model (base station fault prediction model) in the cloud, generating lightweight models suitable for edge computing power through model compression techniques (quantization, pruning), while receiving inference data uploaded by edge Agents (e.g., “base station fault cases detected at the edge”) for cloud model iteration optimization.
Large Model Agent Role: It manages the lifecycle of models at edge nodes (deployment, updates, uninstallation). When the edge detects that “the local model accuracy drops below 85%” (due to new unrecognized fault types), it automatically requests an updated model from the cloud MAAS while uploading edge inference data to the cloud to support model iteration.
Telecommunications Business Scenario:
Scenario Description: The telecommunications company needs to deploy a “device fault prediction model” at edge base stations to predict the risk of component failures such as power supplies and antennas in real-time. When the edge model fails to recognize new types of faults (e.g., “signal fluctuations caused by 5G frequency interference”), it needs to optimize through cloud iteration.
Model Used: MAAS calls the Alibaba Cloud Tongyi Qianwen model compression tool (to generate edge lightweight models), and the large model Agent constructs an “edge-cloud collaboration Agent” based on the 360 edge collaboration model.
Role: MAAS trains the base station fault prediction model in the cloud (accuracy 92%), compressing it from 5GB to 500MB for deployment at edge base stations;
When a certain edge base station Agent detects a “new type of 5G frequency interference fault” (not recognized by the local model, accuracy drops to 82%), it automatically uploads the characteristic data of that fault to the cloud. MAAS iterates the model based on the new data (accuracy improves to 94%), and the Agent receives the updated model and redeploys it, completing the model update for all edge nodes within a week, achieving a new fault recognition rate of 90%.
2.3 Dynamic Optimization of Edge Resources
MAAS Role: It calls resource prediction models to analyze the historical computing power consumption of telecommunications edge nodes (e.g., “Inference tasks account for 60% during peak hours (18:00-22:00)”) and business demands (e.g., “User access log processing has a higher priority than historical data statistics”), generating resource scheduling rules (e.g., “Allocate 70% CPU resources for inference tasks during peak hours”).
Large Model Agent Role: Based on MAAS’s scheduling rules, it monitors the CPU, memory, and bandwidth usage of edge nodes in real-time, dynamically adjusting task resource allocation (e.g., “When CPU usage exceeds 85%, pause low-priority historical data statistics tasks”). When edge nodes fail, it automatically migrates tasks to neighboring edge nodes to ensure business continuity.
Telecommunications Business Scenario:
Scenario Description: A certain edge node of the telecommunications company (covering a commercial area) needs to handle both “real-time analysis of user access” (high priority) and “daily base station log statistics” (low priority) simultaneously from 18:00 to 22:00 on weekdays, avoiding low-priority tasks from occupying resources and causing delays in real-time analysis.
Model Used: MAAS calls the Baidu Wenxin Yiyan resource prediction model (to generate time-based scheduling rules), and the large model Agent constructs an “edge resource optimization Agent” based on iFLYTEK’s Spark resource scheduling model.
Role: MAAS analyzes the resource data of that node over a month, generating rules: “From 18:00 to 22:00, allocate 70% CPU/memory for real-time analysis, pause log statistics tasks; after 22:00, allocate 50% resources for log statistics”;
The Agent dynamically schedules resources according to the rules, keeping the CPU usage of real-time analysis tasks during peak hours stable below 65%, reducing latency from 200ms before optimization to 80ms, and automatically executing low-priority tasks after 22:00, improving resource utilization by 40%.
3. The Collaborative Logic of MAAS and Large Model Agent and Their Value in the Telecommunications Industry
Collaborative Relationship: MAAS serves as the “intelligent capability foundation,” responsible for transforming business needs and data characteristics into standardized rules and lightweight models; the large model Agent acts as the “automation execution carrier,” responsible for implementing rules in global scheduling of data weaving and local decision-making in edge computing. The combination of the two achieves an upgrade from “capable of processing” to “capable of optimizing” and from “static deployment” to “dynamic adaptation.”
Core Value in the Telecommunications Industry:
Improved Data Weaving Efficiency: Data discovery, integration, and access efficiency improved by 60%-90%, with cross-system data association accuracy exceeding 95%, supporting collaborative analysis across multiple departments in telecommunications;
Enhanced Edge Computing Response: Edge data processing latency reduced from seconds to milliseconds, with anomaly handling efficiency improved by 70%, meeting the real-time requirements of telecommunications networks;
Resource Cost Optimization: Resource utilization at edge nodes increased by 30%-50%, and the cloud model iteration cycle shortened by 50%, reducing computing power and operational costs.
Through the deep integration of MAAS and the large model Agent, telecommunications companies can build a “globally interconnected, real-time responsive” data processing system, supporting core business objectives such as 5G network optimization and enhanced user experience.