Experience

Senior Staff Machine Learning Engineer Nov 2022 - Present

Developed machine learning models for robotic control, including data processing pipelines for training control models such as Diffusion Policy and ACT models.
Designed scalable data processing and training pipelines for classification tasks like double pick detection and item recognition. Collaborated with an inventory recognition team to reduce double picks, achieving an 80% recall rate with minimal impact on system throughput.
Built A/B testing analysis tools and advocated for the integration of performance monitoring tools.
Partnered with cross-functional teams to implement ML training pipelines. Mentored colleagues and interns on machine learning tools.

Created the company’s first robotic grasping ML components, leveraging segmentation, multi-outcome prediction, and classical image processing techniques. Achieved grasp success rates exceeding 95%.
Optimized and fine-tuned robotic systems in production environments across four product lines, enhancing performance and reliability.
Developed item identification and tracking models using SKU-level representation embeddings, reducing wrong-item grasping by over 80%.
Promoted cloud-based solutions for data processing (Vertex AI pipelines), model training (Ray), and experimentation monitoring (ClearML). Also developed in-house A/B testing tools.

Led projects analyzing electronic medical records and health system activity data for prediction and policy analysis. For example, used a Bayesian model of HPV-triggered cervical cancer and public data from Kenya to estimate disease prevalence across provinces.
Conducted neuroscience research on functional connectivity using EEG, fMRI, and neurofeedback. Focused on inferring fMRI-like information from EEG and studying brain activation dynamics during emotional changes.
Advised PhD and postdoc interns and taught internal machine learning courses.
Developed ML tools for anomaly detection, clustering, and prediction in neuroscience and healthcare datasets.

Designed machine learning algorithms for a bank’s fraud detection system, reducing false positives by 50% while maintaining alert accuracy. The system used a random forest model with customized feature vectors representing recent customer activities.
Created anomaly detection methods for large system logs in a mainframe monitoring solution. The system identified anomalies by learning message distribution statistics and appearance relationships.
Provided guidance on integrating research into IBM product offerings.
Programmed in Matlab and C++.

Developed rapid prototypes for helicopter “glass cockpit” upgrades. Allowing for testing of system specifications prior to firmware implementation.