About
Senior Software Engineer · MinnesotaSoftware engineer with 10+ years building systems where failures have real consequences: FDA-regulated implants delivering electrical current to patients' brains, structured loan portfolios worth billions, and research infrastructure for experimental cancer data.
How I Work
I learn unfamiliar domains quickly, find where teams are stuck, and automate or eliminate the pain points to accelerate output. I work across the full lifecycle — discovery, design, implementation, testing, deployment, and support — and try to raise team standards along the way, not just ship my own code. I care about code that can be read, maintained, reviewed, and safely evolved by both humans and AI-assisted tooling. If you're solving hard engineering problems, I'd like to hear about them.
Education
- U. of Minnesota
- B.S. Computer ScienceSoftware & Data Systems · 2012–2014
- Bethany Lutheran
- B.A. Physical Sciences2009–2014
Netrias
Jan 2024 - PresentSenior Software Engineer at Netrias, a biomedical AI company, building data harmonization and research infrastructure for ARPA-H's Biomedical Data Fabric program. The work centers on standardizing cancer research metadata so that clinical cohorts from different institutions can be pooled and compared.
Projects
Data Chord
- Architected and shipped a five-stage human-in-the-loop harmonization web application from blank repo to production AWS — slated for GovCloud integration for production government use
- Designed the product UX including a documented triadic-color design system, custom vanilla-JS combobox, and a virtualized in-card row context popup
- Built the hosted stack in OpenTofu (ALB + ECS Fargate + Cognito + S3 + ECR + CodeBuild) with CloudWatch alarms and idempotent staging/prod deploy scripts
Harmonization Inference Pipeline
- Cut inference end-to-end runtime from ~18 minutes to ~30 seconds warmed-up by re-platforming onto a Step Functions + Lambda + SQS distributed-map architecture capable of 10,000 parallel child executions
- Designed a deduplication-first topology that eliminates ~99% of harmonization compute on representative inputs
- Serves the model behind the team's 96% in-dictionary accuracy across 691,000+ cancer vocabulary terms (NCIt, caDSR, GDC, ICD-O3, MedDRA)
Model Training Pipeline
- Identified the need and built a Slack-first ML training orchestrator on Lambda + Step Functions + Coiled with stage-aware error classification and inline observability links on every notification
- Replaced a patience-counter early stopping (silently reset by micro-improvements) with a sliding-window approach, saving ~3 days of compute per training run
- Drove the move from single-GPU to multi-machine NCCL distributed training: 92%/14-epoch on g4dn.12xlarge vs 88.4%/6.82-epoch baseline (~40% runtime reduction, ~2x epoch throughput)
Agentic Harmonization
- Identified an opportunity and built a standalone agentic harmonization library — a Bedrock LLM-agent pipeline that matched zero-shot performance on small standard sets and outperformed it as the standard count grew
CDE Recommendation Service
- Built and deployed a service that recommends, for each user-supplied data column, the common data elements (CDEs) best suited to harmonize against
- Replaced a 9.6 GB Docker pipeline (5 Lambdas, Step Functions, ECR, 145 packages) with a single zip-deployed OpenAI-backed Lambda using 4 runtime dependencies, with no functional regression
netrias_client SDK
- Authored and published the Python SDK on PyPI; adopted as the integration contract by every downstream Netrias service
- Established canonical typed wire shapes with positional-parity invariants from request through manifest, eliminating a class of silent data-loss bugs on duplicate CSV column headers
Notebook to Application
- Authored nb2app, an internal tool that converts arbitrary Jupyter notebooks into multi-arch reproducible Docker images with auto-generated typed GUIs
- Production packaging mechanism for nine CARES (CBRN-Core) customer deliverables shipped on schedule; also supports the RNASSIST bioinformatics suite for alcohol-related gene expression research
Engineering Standards & Proposals
- Founded the Netrias engineering standards repo with orientation, workflow, and repo-standards docs and a tooling compliance matrix
- Contributed software architecture and cost estimation to BD proposals including PANTHEON (awarded) and ASSIST II
ZAIS Group
Jul 2022 - Jan 2024Software Engineer at ZAIS Group, a structured finance firm managing $4.6B in CLO assets. Built data pipelines and search tools for the Data Analytics team, reconciling inconsistent vendor feeds into reliable loan-level metrics.
Projects
CLO Data Pipeline
- Architected a configurable Python/Airflow pipeline that reconciles inconsistent feeds from multiple loan-data vendors, selects the best available value at the field level, and produces reliable loan-level metrics across $4.6B in managed assets
Embedded Search
- Built embedding-based document search across thousands of financial disclosures, cutting analyst research time by 10-100x
CLO Financial Model
- Brought back as an external consultant to solve a scaling problem on an existing portfolio risk model — migrated the desktop tool to cloud infrastructure with horizontal scaling, turning overnight runs into ~two-minute results (~250x speedup)
Medtronic
Jan 2015 - Jun 2022Software Engineer at Medtronic for seven years, building FDA-regulated Android therapy control software for deep brain stimulation (DBS) implants. Deep brain stimulation is an FDA-approved therapy for Parkinson's disease, essential tremor, dystonia, and epilepsy that works by delivering targeted electrical pulses to precise brain structures. The Percept PC system—FDA-cleared in 2020—was the first to sense and record brain signals during therapy, enabling adaptive, personalized treatment. It has since been recognized as a TIME Best Invention and validated in JAMA Neurology.
Projects
Percept PC
- Built Android therapy control system for the first-ever sensing-enabled DBS platform; part of the team that delivered a product now implanted in 40,000+ patients
- Led therapy control software team across onshore and offshore developers -- teaching system behavior, conducting code reviews, and clarifying requirements -- while writing core stimulation control features for a device treating Parkinson's, essential tremor, dystonia, epilepsy, and OCD
- Owned application architecture end-to-end -- from the user-interaction layer through the event system, telemetry modules, and device telemetry formation
- Partnered with neurologists and nurses on UI design and clinical workflows
Prototype
- Built the Android clinician app for a multi-system deep brain stimulation prototype; successful proof-of-concept avoided $50M+ in alternative development costs
- Led cross-team performance characterization, driving a 3x system speedup
Activa
- Shipped Android therapy control software for deep brain stimulation implants now serving 175,000+ patients, against a hard external FDA submission deadline
- Redesigned the telemetry layer to improve error handling and accelerate team development velocity, and steered the application toward a more testable architecture so the team could meet required coverage and verification expectations
- Built a custom dependency-injection framework in-house to eliminate SOUP compliance risk around software provenance, stability, and long-term maintainability
Symptom Tracker
- Built core symptom-tracking functionality and bidirectional data sync for a new patient platform, including REST API and synchronization protocol for mobile<->backend data exchange
- Introduced Kotlin to the department through a production pilot, then taught it to peers across teams -- the pilot influenced adoption across multiple multi-million dollar projects
- Wrote department-wide Java/Kotlin coding standards focused on patient safety, code clarity, and long-term maintainability