The Challenge: Financial Impact of Latent Defects
In the automotive wholesale industry, one of the most significant and recurring operational costs stems from post-sale vehicle arbitration disputes. These disputes arise when a buyer discovers a mechanical condition after the transaction that was not accurately represented at the time of sale.
Among all vehicle systems, drivetrain-related issues are the primary driver of these disputes. Because these faults are often latent during a standard inspection and only manifest under specific driving conditions, they are exceptionally difficult to identify without a more sophisticated, data-driven approach.
The core of the solution was a purpose-built data capture pipeline integrated directly into the existing vehicle inspection app. This system enriches the standard workflow with real-time diagnostic telemetry, introducing a new layer of synchronized acoustic and ECU data without disrupting the inspector’s process or adding operational friction at scale.
By systematically capturing these signals across inspections, the solution generates the structured, high-quality dataset required to train predictive models capable of identifying drivetrain-related risks—extending the value of each inspection beyond immediate diagnostics.
Multimodal Data Capture for Model Training
To capture signals associated with drivetrain behavior, we developed a system that captures a rich signal set by aligning two distinct data streams in real-time:
Acoustic & ECU Integration: While high-fidelity audio captures the engine running at defined RPM ranges and the transmission shifting between gears, a OBD-II adapter simultaneously streams live ECU parameters—including engine RPM, throttle position, and engine load.
Drivetrain Correlation: By synchronizing these inputs, the system provides data that correlates engine sounds directly with internal drivetrain behavior, capturing signals that are not available in standard inspections.
Engineering for Scalable Data Collection
To guarantee the solution worked across thousands of vehicle makes and models, we implemented two key strategies:
ECU Emulation for Accelerated Development: We integrated an ECU emulator throughout development and QA to simulate vehicle signals on demand. This decoupled iteration speed from vehicle availability, significantly accelerating the development cycle by enabling engineers to reproduce edge cases, validate parser logic, and run regression tests without booking bench time on a physical vehicle.
Real-Vehicle Validation: Every milestone was validated against real vehicles to confirm emulator-based results and surface edge cases that only appear in production conditions — including ECU-specific timing variance, intermittent connection loss, and manufacturer-specific PID behavior. This two-tier testing strategy (emulator for velocity, real devices for fidelity) ensured the system performed reliably across the diverse hardware landscape it would encounter in the field.
Predictive Data Foundation: The solution enables the collection of data at scale, building the foundational dataset required to train predictive models capable of identifying drivetrain risk signals with high confidence.
Architecture for Reliable Data Capture
To handle high-fidelity telemetry in the background without impacting app performance, we implemented a robust engineering stack designed for stability and real-time processing.
Custom Native Bridge: We developed a bridge using Expo’s Module API to expose the OBDII devices SDK to the React Native layer, ensuring seamless communication between the hardware and the cross-platform UI.
Modern Swift Concurrency: We leveraged Swift's native concurrency model to stream OBDII telemetry asynchronously, guaranteeing that high-frequency hardware events never block the UI thread and that resources are deterministically released when the connection closes.
Clean Architecture on the Native Layer: The iOS module was built following Clean Architecture and SOLID principles, enforcing a strict separation between data acquisition, domain logic, and UI state. This separation is critical for a module that must run reliably in the background and remain testable in isolation.
Core Technologies: The solution combines React Native for rapid iteration with Swift, Objective-C, and C++ for high-performance hardware integration.
Scalable data pipeline
Synchronized collection of high-fidelity engine audio and live ECU data.
Seamless integration
Telemetry capture was added without disrupting the inspector’s workflow or impacting app performance.
New diagnostic visibility
The system enabled capture of latent drivetrain signals across real inspection conditions.
Reliable implementation
The native hardware integration was built using clean architecture principles, making the feature testable, maintainable, and resilient in the field.
Business impact
Because drivetrain-related issues are a major contributor to post-sale arbitration, the solution directly supports ACV’s ability to reduce arbitration exposure by improving early visibility into latent mechanical risks.
Key Takeaway
The project showed that predictive drivetrain intelligence depends on both strong engineering architecture and real-world validation. ECU emulators were valuable for accelerating development and regression testing, but they could not replace field testing with real vehicles, where manufacturer-specific behavior, timing variance, and connection instability had to be validated under production-like conditions. By applying clean code principles and separating hardware acquisition, domain logic, and UI state, the team delivered a reliable, testable foundation for scalable predictive data collection.
