Mike Powell Email and Phone Number

Using Computer Vision / AI to Improve Dental Healthcare

* Increased prediction accuracy in by 125% (ρ:20%→45%) over previous best-in-class models by developing a novel encoding strategy for ingesting billions of virus measurements into a supervised deep-learning neural network.* Identified 30 (of 100) design seeds as failure points by automating metrics and reporting, which surfaced these failures. It is projected that eliminating these seeds will give the single largest improvement in the next design run.* Decreased operation time by 75% for the main design pipeline by overhauling interface and streamlining components. This additionally enabled: (1) better end-to-end testing and (2) clearer insight into design decisions.* Reduced compute cost by 90% for the performance-forecasting pipeline by (1) performing exploratory data analysis and experimentation, (2) productionalizing findings, and (3) eliminating redundancies.* Built community, created roadmaps, and established best practices by mentoring coworkers on contemporary standards and by leading code reviews. Collaboratively weighed design trade-offs for long-term projects.

* Forecasted cancer remission from CT scans (AUC-ROC 0.92 / PR-AUC 0.94) by (1) rapidly prototyping new computer vision capabilities, (2) creating cloud-based production pipelines, and (3) automating metrics and reporting.* Predicted immunotherapy risk (CI 0.74) by prototyping and deploying semi-supervised clustering and classification algorithms that identified risk factors and stratified patients into cohorts with 10%-600% the background risk.* Reduced container build time by 95% by (1) implementing team-wide continuous-integration, (2) adapting SOTA tools for dependency management, and (3) establishing team standards and expectations.* Accelerated team development velocity by (1) surfacing metrics with automated reporting, (2) creating end-to-end analysis pipelines, (3) developing APIs for cloud ops, and (4) establishing common testing/documentation standards.

* Accelerated call center operations by 78% by developing and deploying NLP / NLU / ML algorithms and RESTful APIs that auto-populated helpdesk tickets. Full deployment was projected by the client to save $4,700,000 annually.* Automatically solved 73% of client queries (95% human-concordance) by (1) developing text-mining algorithms that extracted solutions from historical databases and (2) assigning solutions via a classifier.* Maintained and upgraded the build and deployment pipeline by automating (1) retraining, (2) evaluation, and (3) reporting while adapting and extending SOTA versioning and package management tools for multi-language projects.

* Predicted Alzheimer’s progression from MRIs (90% TP / 85% TN) by mining multimodal medical databases and mathematically generalizing SOTA M-theory computer vision algorithms to operate in three-dimensional space.