- Manage a team of 4 to 6 data scientists developing algorithms to non-invasively predict blood biomarkers using novel silicon photonics wearable medical device and SWIR optical sensors.- Design ML lifecycle to manage data streams, feature engineering and storage, model inference, model monitoring, shadow testing, detect anomalies, and regenerate stale models.- Build ML Ops using Pulumi IaC, MetaFlow, MLFlow, DVC, AWS services (Firehose, Lambda, API, TimeStream, Batch, S3, DynamoDB, CloudWatch, Step Fn), and write Python SDKs to manage data and pipelines.- Lead ML development in human studies. Collect high frequency timeseries data, perform exploratory data analysis and data visualization, signal processing and feature engineering, train models using PyTorch, XGBoost, and Scikit Learn and continuously deploy into cloud.- Management responsibilities include developing ML R&D roadmap, performance evaluations, technical mentorship, managing conflicts, project timelines, and risks.- Collaborate with domain experts and stakeholders from large MedTech companies, consumer health companies, and research organizations. Cross-functional collaboration with business teams to develop product and engineering requirements.- Document software and algorithms in accordance to IEC 62304, ISO 13485, ISO 9000, and ISO 26262 to meet standards for the medical device industry.

I managed an analytics program for an end-to-end carbon intensity reporting for all Amazon Transportation Services in North America and Europe. This was used to help find blindspots in carbon emissions and track metrics to reach net-zero carbon by 2040.

I worked on several R&D projects in the Janssen DSAI organization. One of the large projects was to build a clinical trial screening algorithm using digital pathology. I trained a semi-supervised deep learning model using whole-slide images from internal clinical trials. I then designed and built a web application to serve the model and create visualizations to guide researchers. We validated this with pathologists in the organization to further fine-tune the model.

I studied a novel deep neural architecture to model immunotherapy survival using negative binomial likelihood.

Building a versatile bioinformatics web tool for epidemiologists, physicians, and health investigators.