Joe Shull

About Me

There are few things that can compete with the feeling of turning data into real information and business action. For me, it's an abstraction of the art process: take raw materials, understand the materials, and refine them into something beautiful *and* useful.

I'm currently pursuing a career in data-science, and am interested in opportunities requiring machine-learning and big-data engineering.

If you're interested in chatting about data or tech, drop me a line and let's be friends!

Experience

Lakewood Steel

Founder

Lead fabricator, bid estimator, product designer.
• Delivered $100k in commercial steel fabrication for well-known brands: Xcel Energy, Zoe’s Kitchen,etc
• Designed the signature architectural lights featured in the Leesburg, FL city gateway

Anadarko Petroleum Corporation

Financial Analyst

Capex Planner/Forecaster for a $2B Oil and Gas asset north of the Denver Area
• Budgeted and forecasted asset operations including operated development, non-operated activity, infrastructure build-out, and other non-drill activities
•. Analyzed capital spend and delivered frequent presentations to executives regarding capital trends and events.
•. Built robust modeling tools in Essbase and Excel to achieve < 3% forecast variances.

Education

Galvanize

2018

Data Science Fellow

A 13-week immersive with 700+ hours of coding, weekly Case Studies, and 3 capstones. Python-based curriculum focused on machine learning and best practices in statistical analysis, including frequentist and Bayesian methods. Utilizes regression, classification, and clustering to model real-world structured and unstructured data. Explores NLP and Deep Learning techniques, in addition to Spark on AWS.

University of Colorado

2007 - 2011

Bachelor of Science: Accounting and Information Systems

Economics, Minor

Projects

Two-Sigma / Kaggle : Using News To Predict Stock Prices

Designed and trained a Gradient Boosting Machine to predict short-term stock returns
• Achieved +0.68 mean-variance on market-adjusted returns over 18+ months of predictions.
• Model performance currently places top-18% in competition.

View Project

What The Fish

An exploration in image classification using Logistic Regression and CNNs
• LR: Scraped, reformatted, and converted 2000 images for classification with SKLearn’s Logistic Regression: achieved AUC of 79% on binary problem.
• CNN: Trained and evaluated on a hand-curated 25,000 image dataset: Achieved 79% Top-3 Accuracy on 60+ species
• Coming soon: What The Fish App, a CNN serving predictions to a mobile front-end via Flask.