Aspiring machine learning engineer exploring neural networks, local inference, and the systems that turn models into capabilities.

Email · LinkedIn · Seattle, Washington

Machine learning fascinates me most when it moves beyond a notebook. I want to understand what the model is learning, how architectures and weights shape its behavior, what it takes to run inference locally or behind a service, and how data pipelines, evaluation, and orchestration turn a model into something a product can actually do.

That is the direction behind my work: computer vision experiments, reproducible data science, model-assisted content pipelines, ML education through games, and the software foundations those systems need. I am still becoming a stronger software engineer, but the reason I care about the engineering is simple: ambitious ML only becomes useful when the surrounding system can carry it.

I am interested in every layer—and especially the handoffs between them. A metric can tell me whether an experiment worked; the full system determines whether that learning becomes a reliable capability.

Traffic Sign ResNet50 · modeling + inference TensorFlow · transfer learning · fine-tuning · class weights · TTA Removed a 99.92% physical-sign group overlap from the original evaluation, then reached 92.90% accuracy and 91.29% macro F1 on 7,830 images from 261 unseen sign groups.	Data Science Portfolio · data + evaluation pandas · scikit-learn · feature boundaries · reproducible reporting Built a tested property-age classifier over 22,913 records with a fixed holdout, explicit leakage exclusions, generated evidence, and a 0.9768 ROC-AUC.
Gradientfall · generation + orchestration external models · JSON schemas · validation gates · Godot runtime Designed a model-assisted content flow where briefs become generated JSON, enter a non-shipping inbox, pass machine validation and editorial review, and only then become runtime-loaded game content.	Neural Quest · ML as an interactive experience features · classical models · deep learning · representations · attention Turns a 20-world ML curriculum into places, tutors, experiments, encounters, and boss checks so the concepts become interactions instead of vocabulary lists.
Emvera · secure application systems Django · Plaid Sandbox · PostgreSQL · TOTP · Docker · 205 tests Builds the kind of secure, owner-scoped, observable application foundation that future data and model capabilities should inherit rather than work around.	Pizza Sales Analysis · data into decisions pandas · validation · aggregation · decision-ready visualization Transforms 48,620 transaction lines into reproducible views of demand, staffing pressure, category performance, and product opportunity.

• How far can useful inference move onto local hardware, and what changes when privacy, latency, memory, and compute become hard constraints?
• How do data quality, representations, architecture, learned weights, and evaluation interact in a model's real behavior?
• How should orchestration combine models, tools, deterministic code, validation, and feedback loops?
• What turns a successful model experiment into a capability that people can actually notice and use?

I am working toward the ability to take a model from raw data all the way to a useful, observable capability. The next stretch of that path is deeper neural-network and representation-learning intuition, more hands-on work with local models and inference tradeoffs, stronger end-to-end data/model pipelines, and the production engineering needed to make those systems dependable.

If you want to talk about applied ML, local inference, model systems, or building unusual AI-powered products, reach me at [email protected].