Event Recap | Carlinx Senior Product Director Liu Yixuan Unpacks How Domain-Specific Agents Close the Data Loop

When a vehicle breaks down, how quickly can an operator detect the issue? And how quickly can the right course of action be determined? In the AI era, the answer often lies in having a “flight director” capable of transforming collected data into diagnostic insights and turning those insights into action.

Recently, the 2026 "Equipment Powerhouse: Entering GAC" concluded successfully. Mr. Liu Yixuan delivered a presentation on the theme of “High-Value Data Application Practices for Intelligent Connected Vehicles.” Grounded in real business scenarios, he broke down the complete chain, from vehicle-side perception of high-value data to a vertical-domain agent that drives the business loop to completion.

The following is excerpted from his speech:

Starting Point: Get the Math Right on Data Acquisition

Data is a critical enabler for an agent's operation. Yet under traditional acquisition models, the cost equation has long been difficult to balance.

Carlinx's answer is the Vehicle-Cloud Intergrated Platform, which turns the vehicle itself into the "first processing layer" for data. Through edge computing, data is filtered, processed, and prioritized before it ever reaches the cloud, reducing costs and increasing data value at the source.

Trigger: AI-Driven Conversational Diagnosis

Data acquisition is a long-term operational record. With the "fuel" of data in place, the real value lies in putting it to work.

Carlinx's answer is the Vehicle-Cloud Intergrated Platform, which turns the vehicle itself into the "first processing layer" for data. Through edge computing, data is filtered, processed, and prioritized before it ever reaches the cloud, reducing costs and increasing data value at the source.

                  Anomaly Analysis                                        Conclusion Generation

Traditional remote diagnosis is often limited by incomplete data coverage, poor scalability, and conclusions that are difficult to reuse. The Hyper Agent start by expanding  the perception entry. Multi-source data, including vehicle logs, CAN signals, SOA messages, and DTCs, together with natural language and user voice interactions, all flow into the diagnostic system.

When a vehicle anomaly occurs or the user actively reports a fault, the agent immediately initiates the diagnostic workflow, correlating multi-source signals with operating conditions and historical cases. When the abailable evidence is insufficient, it automatically triggers additional data acquisition task to fill the gaps and complete the diagnostic picture.

Based on multi-dimensional cross-validation, the Hyper Agent can rapidly pinpoint the fault location and root cause, while generating actionable remediation guideline. This enables a new form of "conversational diagnosis."

Closed Loop: Knowledge Accumulation and Self-Evolution

The Hyper Agent is never a one-time delivered product, but a capability curve that keeps rising.

With every diagnosis, the resulting insights are transformed into reusable knowledge assets that can be retrieved by users, paired with visualized text-and-graphics repair guidance, or used to tailor training programs for dynamic scenarios. The diagnostic model continuously optimizes and evolves through ongoing evaluation. From fault perception and analysis to experience accumulation and feedback, every cycle of the data loop makes the system a little smarter.

Actual operational data shows the diagnostic-and-remediation agent now covers nearly a thousand service stations and nearly ten thousand repair technicians, cutting single-vehicle diagnosis time by 40% and annual operating costs by 30%.

It is precisely this capability, already proven in real-world scenarios and corroborated by operational data, that has drawn widespread attention to the Hyper Agent.