Senior AI/ML engineer and team lead

Michael S. Davinroy

I lead engineering teams and product pilots. My technical work includes machine learning, model analysis, security research, robotics simulation, and deployment.

MS in Computer Science, Northeastern University. BA in Computer Science with Honors and an Honors minor in Mathematics, Swarthmore College.

Work

I help shape new programs, lead engineers, run customer pilots, and move successful work into regular use.

Business development: I help shape new work through technical proposals, customer workshops, and planning with partners.
Engineering leadership: I set technical direction, assign work, mentor engineers, and support hiring and onboarding.
Product pilots: I run pilots with customers, gather feedback from users, and turn it into product requirements and launch decisions. This has included work with news organizations and international child-safety teams.
Deployment: I plan the data, computing infrastructure, user interfaces, and day-to-day workflow needed to put a model into use.

Approach

I begin with a question we can test and a baseline we can measure. I examine failure cases before deciding whether a more complex model is worth the cost.

I plan for real operating conditions early. Feedback from a pilot can change the model, the workflow, or the product plan.

Technical

Core methods

Machine learning and model analysis: I train and adapt transformer and self-supervised vision models with PyTorch and PyTorch Lightning. I design test sets and study calibration, ROC/PR curves, shifts in input data, false positives, and false negatives.
Systems: I work on projects spanning from low-level systems research with QEMU/KVM to multi-GPU training and model analysis on AWS EC2/S3.
Visualization: I use techniques and technologies including PCA, t-SNE, Plotly, and Dash to examine model behavior and explain the results.

Applied research

Security research: I study adversarial attacks on ML models used for binary analysis and cache side-channel attacks on ML frameworks.
Robotics simulation: I develop and evaluate legged-robot locomotion in MuJoCo and Isaac Sim, including entangling vegetation and wet, muddy, or sandy terrain.