Software Developer in Corvallis, OR, United States

In past roles, a large part of my day-to-day work centered around creating processes that present data in a useful form. In these projects, the end form might be a simple report, a real-time dashboard, or an API or process that exposes data for further analysis in Python, R, or BI software.

No matter what the end product is, the process starts with understanding the value the business expects to extract from data as well as any expected challenges or known deficiencies in the data. As a second step, I establish access to the data in R or Python and start exploring. I'm looking to see how complete the data is, whether there appear to be errors, and most importantly, am looking to see that the data shows the trends or relationships expected by the business.

The final step is to set up reports, dashboards, or APIs that expose the data. Usually, this will involve creating intermediate processes to clean, transform or process the data as well as setting up monitoring to make sure that the intermediate processes continue to function properly in the background.

I used and extended the PyQtGraph library to develop a flexible plotting tool capable of interactively displaying ~100,000 data points. It allows for seamless zoom from minute to microsecond timescale and incorporates data augmentation features such as the ability to click on points to display additional information.

I developed code to create highly performant custom data loaders to load arbitrary binary data to Pandas/NumPy in Python. The techniques developed also support high performance loading of irregularly structured text data.

I've competed in a few Kaggle contests and done well in all of them. I finished individually in the top 5% globally in the 2019 Data Science Bowl, and my team and I finished in the top 10% in the TensorFlow 2.0 Question Answering contest. The 2019 Data Science Bowl involved semi-structured data and traditional machine learning techniques such as gradient boosted trees, bagging, and feature selection and engineering. In the TensorFlow 2.0 contest, my team and I applied the Bert NLP model to a question answering dataset.