• Researched and developed on-device foundation model for the sensor modality, achieving 7-15% KPI improvement on downstream tasks (prod algos) with 30-50% less compute/memory.

• Researched and developed Advanced Running Dynamics ML algorithms featured on Pixel and Android devices, demonstrating 10% better performance to Apple & Garmin by third-party testing.

• Designed and implemented a modular, optimized training pipeline for motion algorithms, achieving 70-100x faster runtime. Enabled rapid hyperparameter search, resulting in the most widely adopted motion algorithm training pipeline in the Product Area.

Building few-shot sequential models for gesture recognition.

• Developing novel few-shot semi-supervised sequential learning algorithms for sensor-based data streams for wearable devices and achieving state-of-the-art performance with less than 50 examples

• Devising comprehensive testing mechanism to ensure real-world replicability including developing real-time WearOS / Android app, sequential error modeling, false positive testing on numerous datasets

• Working with key stakeholders to identify success metrics, leading data collection efforts, spearheading cross-team collaboration, managing tasks, and expectations with senior engineers

Using unsupervised techniques such as Deep Generative Models, Reinforcement Learning, and

Probabilistic Graphical Models to detect and remove erroneous data from large DBs at Google. I developed a novel generative adversarial imputation network designed to detect and recommend

corrections to erroneous data fields with much better results than current industry standard solutions. These results were documented in an internal white paper where we discussed the problem, approach, and compared key metrics and results. This was done through a multi-team collaboration where we worked with key stakeholders across multiple teams to understand their problems, collaborate to find a solution that addresses their key concerns, and learn key insights about development at needed scale.