Developed predictive models that help farmers make planting and field management decisions. Spearheaded the implementation of MLOps best practices and trained data scientists how to develop production-ready machine learning models. - Leads the “Code Quality & Lifecycle” team to establish software and ML model development best practices throughout the department. - Developed and led training sessions to teach data scientists how to structure high quality codebases, write unit tests, automate data pipelines & model training workflows, and maintain models in a production environment. - Refactored legacy ML model codebases, reducing tech debt and cutting experiment runtime from several weeks to less than a day. - Created an internal Python package to enable scientists to automate the creation and documentation of tables in Google BigQuery.

Focused on improving the reliability of in-app location services, predicting driver on-time performance, and optimizing workflows for operations specialists.- Deployed a classification model to predict the likelihood that a driver will arrive on-time to a facility, improving the late detection rate by 10%.- Developed a regression model to predict when a driver will arrive at a facility, improving predicted ETA accuracy by 25%.- Implemented a heuristic model to determine whether we’re tracking the correct driver, reducing incorrect driver tracking by 80%.

Focused on extracting information from engineering drawings to provide comprehensive part lists for mechanics on the shop floor.- Trained a Faster-RCNN object detection model to locate and identify different classes of symbols in engineering drawings.- Applied classic image processing techniques and open-source optical character recognition models to interpret text and identify parts in engineering drawings.- Implemented and optimized a search engine in Elasticsearch to find maintenance reports relevant to problems an engineer was currently working on.

- Developed Gefilter Fish, a Chrome extension for Amazon customers that identifies unique topics from reviews using natural language processing (NLP). Extension reduced redundancy in review topics by more than 15% compared to current feature on Amazon.- Aggregated over 1.7M reviews into a PostgreSQL database (python & pandas). Preprocessed unstructured text with NLP (lemmatization & tf-idf; nltk & sklearn) to identify unique topics (non-negative matrix factorization).- Assessed overall customer perception of a review topic using the VADER sentiment analysis tool.- Scraped product pages to compare generated topics with current feature on Amazon (selenium, Javascript).- Built a Chrome extension (Javascript) to run the model on an AWS server and modify product pages. Extension displays topics and summarized reviews focused on that topic.

Research:- Developed software to predict the properties of small molecules and how they interact with X-ray laser light. - Devised a new algorithm to efficiently describe how molecules interact with X-ray lasers. This technique was up to 1000 times faster than previous approaches and is now available in the commercial software package Gaussian16 (Fortran77).- Implemented several optimization algorithms to converge quantum mechanical wave functions in ∼10 times fewer steps than earlier optimization techniques (C++).- Collaborated with a team of 10 experimental researchers to model how muscular proteins bind and release small molecules and how this process can be tracked with X-ray lasers.Outreach:- Wrote proposals to fund the purchase of 96 compute nodes and started a High Performance Computing Club (HPCC) to manage this resource. In less than 2 years, the club has grown to >300 students from >20 different departments.- Worked as an account manager and as technical support for the HPCC, handling many software packages such as Gaussian, ORCA, MATLAB, and others.

Worked closely with a team of 5 software developers to implement an algorithm to evaluate the energy of a molecular system up to 30% faster.

- Developed an automated scheme to determine of the absolute configuration of organic compounds using Vibrational Circular Dichroism spectroscopy. - This work became the basis for a laboratory exercise in York College's Physical Chemistry I (CHM 344/345).