Role Description
As an Embedded ML Engineer on the Innovation Team, you will design and manage end-to-end machine learning systems that continuously enhance in-vehicle audio and infotainment experiences. You will transform raw and real-world data into production-ready solutions while ensuring seamless integration, monitoring, and iteration of models deployed on embedded platforms. You will be involved in the full lifecycle from training to inference, collaborating closely with Data Scientists, DSP Engineers, and Audio Experts to design models and audio features.
Your key responsibilities will include:
-
Optimizing model performance on embedded devices and enabling remote updates to deliver continuous improvements based on new data and enhanced features.
-
Establishing a closed-loop system by integrating telemetry, model performance monitoring, and continuous retraining or system enhancements based on real-world usage, while adhering to strict privacy, safety, and reliability standards.
Qualifications
-
6+ years of end-to-end ML development experience, including model training, optimization, deployment, and monitoring.
-
Proficiency in C/C++ programming language for embedded systems.
-
Experience working with audio, image or signal processing domains.
-
Experience building data collection architecture design and overall Data Engineering infrastructure.
-
Experience working with automotive platforms such as Android Automotive OS, QNX, and embedded Linux.
-
Understanding of audio-hardware communication protocols such as TDM and I²S.
-
Experience profiling and optimizing models on SoC or DSPS (Qualcomm, NXP, TI, NVIDIA SHARC, TI).
-
Practical experience deploying models on embedded devices such as ARM Cortex cores, GPUs, NPUs, and DSPs (e.g., SHARC).
-
Experience building MLOps pipelines, including automated training jobs, dataset versioning/switching, model optimization workflows, and unit tests for ML code.
-
Strong understanding of real-time systems execution constraints (scheduling, interrupts, shared-core resource).
-
Strong proficiency in Python and production-grade machine learning using frameworks such as TensorFlow, PyTorch, scikit-learn, NumPy, and Pandas.
-
Experience with MLOps practices and CI/CD automation for ML model delivery.
-
Experience working with data lakes and cloud data platforms (AWS S3, Azure, Databricks, etc.) for storage, processing, and training pipelines.
-
Expertise in edge model optimization techniques including quantization (dynamic and post-training), pruning, model compression, and mixed-precision execution.
-
Working knowledge of privacy-preserving data handling (pseudonymization, anonymization, etc.), data-quality checks, and data-lineage tracking.
-
Strong communication and collaboration skills, with the ability to work effectively in an intercultural and cross-functional team.
-
Proven ability to write unit tests, integration tests, and performance benchmarks for ML models and data pipelines.
-
Ability to work in rapid prototyping environments; self-driven, fast learner, with a strong passion for innovation and problem solving with a strong sense of ownership to reach goals on time.
Requirements
-
Deploy ML models on embedded devices such as ARM processors, GPUs, NPUs, SHARC, and other DSP architectures.
-
Optimize models for edge hardware (e.g., quantization, pruning, distillation) to ensure real-time execution within strict latency, memory, and compute budgets.
-
Collaborate with Data Scientists and DSP Engineers to convert research notebooks into robust, testable training and inference code; define evaluation criteria, acceptance thresholds, and quality guardrails.
-
Build end-to-end ML systems, including telemetry ingestion from vehicles, data validation, feature processing, training, evaluation, packaging, and deployment to production environments.
-
Implement continuous training and continuous deployment (CT/CI/CD for ML) with full traceability of model and version registries, along with safe rollback mechanisms.
-
Partner with Embedded DSP Engineers to ensure real-time execution performance, leveraging hardware accelerators such as SIMD, MMA or NEON units.
-
Estimate and measure model-footprint metrics (CPU utilization, memory usage, latency, etc.).
-
Redesign or restructure model architectures to reduce embedded resource consumption while maintaining similar levels of accuracy and overall model performance.
-
Collaborate on OTA and data-collection strategies to support continuous model improvement while adhering to strict privacy constraints.
-
Contribute to internal documentation, reusable components, and best practices; mentor peers on MLOps workflows and edge optimization techniques.
-
Implement monitoring systems to track model performance, data drift, and system behavior in production, ensuring reliability and continuous improvement based on real-world usage.
-
Provide guidance to less-experienced team members on model deployment strategies and automation of ML pipelines.
Benefits
-
Flexible work environment, allowing for full-time remote work globally for positions that can be performed outside a HARMAN or customer location.
-
Access to employee discounts on world-class products (JBL, HARMAN Kardon, AKG, and more).
-
Extensive training opportunities through our own HARMAN University.
-
Competitive wellness benefits.
-
Tuition reimbursement.
-
“Be Brilliant” employee recognition and rewards program.
-
An inclusive and diverse work environment that fosters and encourages professional and personal development.
