The posting
Build, train, and deploy efficient ML models and inference pipelines for real-world Android devices. Focus on time-series sensor intelligence, lightweight classifiers, and scalable backend learning systems. Key Responsibilities Design, train, and optimize small ML classifiers (XGBoost, RandomForest, SVM, MLP, 1D-CNN, GRU/LSTM when sequence context is required). Build feature extraction pipelines for high-frequency sensor data (100 Hz class problems, 1-sec to multi-sec windows). Ensure models are device-agnostic, orientation-invariant, and robust to sensor noise across a diverse Android hardware mix. Implement real-time inference pipelines that run efficiently on CPU-constrained mobile environments using Android NDK or TFLite-compatible formats. Build backend learning services that can handle thousands of concurrent sessions and continuous model improvement from field data. Own model validation, performance benchmarking, drift analysis, false-positive reduction, and temporal smoothing. Collaborate closely with native C++ and DevOps teams for clean deployment and monitoring. Requirements Strong command of ML for time-series and sensor data. Hands-on experience with lightweight models that run on Android devices. Expertise in XGBoost or tree-based classifiers + feature engineering. Experience converting models to TFLite, ONNX, or similar mobile inference formats. Deep understanding of normalization, interpolation, filtering, debouncing, HMM/majority voting, and classifier calibration. Solid grounding in model evaluation metrics and edge-case handling. Good-to-Have Experience with sensor fusion pipelines, movement feature intelligence, or inertial embeddings. Deployment experience on Kubernetes + cloud monitoring.
