Cygnus Services builds machine learning systems for scientific applications – remote sensing, geospatial analytics, hyperspectral imaging – and the validation frameworks that make their behavior understood, defensible, and ready for production.
ML pipelines from scratch
Architecture selection, training and evaluation infrastructure, experiment tracking, and data versioning – engineered for traceability from the first commit. Docker, GitLab CI/CD, DVC, MLflow.
Know what the model actually learned
Attribution and sensitivity analysis, failure-mode characterization, and performance under distribution shift. Verification built around the pipeline as a first-class stage – not appended after the fact.
Production operation without surprises
AWS Step Functions, container orchestration, CI/CD-integrated verification gates, and full artifact lineage from every result back to run, commit, configuration, and data checksum.
Getting a pipeline started
A structured consultation on moving a model from prototype to production – deployment architecture, container orchestration, CI/CD integration, and monitoring. Delivered as a written deployment plan with a recommended architecture and prioritized next steps.
Harden a pre-existing system
An independent assessment of an existing model or pipeline: what it learned, where it fails, and whether the evidence supports deployment. Delivered in 2–4 weeks as a written technical report with prioritized recommendations.
For pipeline builds, deployment engineering, and longer engagements – use the contact form.
Production MLOps pipeline for methane plume detection on hyperspectral satellite imagery, built for an environmental monitoring organization – from scratch, then wrapped in an embedded explainability suite: attention analysis, integrated gradients, band occlusion, sensitivity analysis, and spatial scene evaluation, with full artifact lineage from every evaluation back to its training run.
Principal Investigator on DARPA's Explainable AI (XAI) and Competency-Aware Machine Learning (CAML) programs – causal explanation methods for deep networks, and human-subjects studies demonstrating that causal explanations corrected operators' mental models and measurably improved joint task performance.
Designer of a proprietary verification and validation framework for enterprise ML pipelines – conditional verification driven by Sobol sensitivity analysis, integrated into CI/CD, with isolated inference endpoints that keep proprietary weights and data private.
Hop Doctor – an iOS application that analyzes bicycle jump technique from video using on-device pose estimation: model integration, scoring methodology, and interface designed, built, and shipped end to end as a complete consumer product.
PhD, Civil & Environmental Engineering
University of Minnesota
BS, Physics & Mathematics
University of Michigan
Causal Learning and Explanation of Deep Neural Networks via Autoencoded Activations · 121 citations
Explainable Artificial Intelligence for Increasing User Trust in Deep Reinforcement Learning Systems · 50 citations
Brittle AI, Causal Confusion, and Bad Mental Models · 35 citations
Describe the system and the goal – replies within two business days.
Or connect on LinkedIn.