- Design, develop, and optimize on-device machine learning models for learning user preferences within the App Store ecosystem.
- Implement, benchmark, and fine-tune deep learning models for efficiency, latency, and power consumption on iOS devices.
- Work with Core ML, Metal, and other Apple ML frameworks to deploy models efficiently and securely.
- Ensure all ML computations remain on-device, preserving user privacy while maintaining model effectiveness.
- Develop and apply Private Federated Learning and Differential Privacy techniques to ensure user data never leaves the device while enabling personalized experiences.
- Optimize models for various iOS devices, ensuring performance across different generations and hardware capabilities.
- Develop tools and techniques for model compression, quantization, pruning, and hardware acceleration to enable real-time inference.
- Stay updated with the latest developments in privacy-preserving AI, on-device learning, and Apple ML advancements.
3+ years in ML engineering with a focus on on-device/mobile ML, preferably iOS.
Strong proficiency in Swift, Objective-C, and Python.
Experience with Core ML, Metal, TensorFlow Lite, PyTorch Mobile, or similar frameworks.
Knowledge of model quantization, pruning, knowledge distillation, and hardware acceleration (e.g., Neural Engine, GPU optimizations).
Familiarity with iOS development tools, Xcode, and app deployment processes.
Experience or familiarity with Private Federated Learning and Differential Privacy is a strong plus.
On-Device Personalization: Experience with learning user preferences without data leaving the device is highly desirable.
Ability to troubleshoot and debug performance bottlenecks in ML models and iOS applications.
Hands-on experience with Private Federated Learning and Differential Privacy.
Experience with Apple Neural Engine optimizations.
Background in embedded systems or edge computing.
Knowledge of real-time inference in resource-constrained environments.
Experience building personalization models that adapt on-device without cloud interaction.