Career & Research

Engineering timeline, research, and operating principles.

A decade of architecting distributed streaming systems, applied machine learning platforms, and statistical inference frameworks for mission-critical operations.

Staff Data Science Engineer Distributed Data Platforms Remote worldwide

Demo content. Career trajectory spanning distributed data platform engineering, applied machine learning systems, and statistical research.

Work

Roles and responsibilities.

The through-line is a steady move toward work that combines product thinking, visual craft, and front-end implementation.

  1. 2022 — Present Current

    Staff Data Science Engineer

    Distributed Intelligence Systems · Remote

    Architecting distributed streaming pipelines, sub-millisecond feature stores, and real-time inference clusters for enterprise clients.

    • Architect real-time streaming feature stores ingesting over 1.8B events/day with Kafka and PySpark
    • Design calibrated ML ensembles for fraud detection and ranking with sub-15ms p99 latency SLA
    • Lead infrastructure migration to Triton Inference Server and Ray Serve, cutting compute spend by 40%
    • Establish automated CI/CD evaluation pipelines preventing data drift and model regressions
  2. 2019 — 2022

    Senior Machine Learning Engineer

    Northline Analytics · Portland, OR

    Led production ML engineering for large-scale predictive telemetry and high-throughput scoring systems.

    • Engineered predictive maintenance models across 50,000 IoT sensors using PyTorch and DuckDB
    • Built company-wide Python feature engineering libraries and standardized dbt data transformation models
    • Implemented rigorous Bayesian A/B testing framework with CUPED variance reduction
    • Mentored four junior and mid-level data engineers and machine learning practitioners
  3. 2017 — 2019

    Quantitative Data Engineer

    Fieldwork Quantitative Data · Seattle, WA

    Engineered real-time Kafka and PySpark data pipelines processing billions of events daily.

    • Built distributed ETL pipelines across Delta Lake and PostgreSQL for quantitative research
    • Automated continuous data quality assertions with Great Expectations and dbt tests
    • Optimized analytical SQL queries reducing analytical execution latency by 60%
    • Collaborated with data scientists to package research prototypes into production Docker containers
  4. 2015 — 2017

    Data Scientist

    Synthetix Labs · Seattle, WA

    Developed statistical models, predictive algorithms, and automated analytics for operational forecasting.

    • Trained gradient boosted decision trees for customer churn and demand prediction
    • Engineered automated feature extraction pipelines in Python and Pandas
    • Authored internal technical documentation on statistical testing and model validation best practices

Education

Where the foundation came from.

2013 — 2015

M.S. in Computer Science (Machine Learning & Distributed Systems)

Pacific Institute of Technology · Portland, OR

Graduate research focused on distributed consensus algorithms, high-dimensional probability, statistical learning theory, and database query optimization.

2009 — 2013

B.S. in Applied Mathematics & Statistics

University of Washington · Seattle, WA

Rigorous foundation in linear algebra, multivariable calculus, stochastic processes, and numerical optimization.

Working principles

What stays consistent.

01

Mathematical rigor meets operational reality

A model that achieves 99% accuracy on a static test set is useless if it degrades silently under real-world non-stationary distribution shifts.

02

Zero training-serving skew

Offline training features and online streaming features must share identical schema contracts, unit tests, and transform definitions.

03

Latency is a core product feature

In production, an inference model that breaches SLA latency bounds is indistinguishable from a total system failure.

04

Continuous observability over blind faith

Live prediction distributions, input feature drift, memory pressure, and p99 percentiles must be monitored with automated circuit breakers.