Projects
Things I've built, contributed to, or am tinkering with.
keegankelsey.io
2026active
This site — built with Astro and Tailwind, with posts imported straight from Notion exports.
Semontos
2026active
Independent project, private repo. A semantic layer over tabular data with a Model Context Protocol server, so agents retrieve data by meaning rather than by column name.
claude-codeprivate-applicationmcp-servercliapiSynthetic Data (insilicodna)
2022active
A Python package for generating synthetic molecular data.
Why create synthetic molecular data when there is an ever-growing body of real data to use? Simply stated, control. For the Engineer or Data Engineer, it is critical to have data when building pipelines. For the Scientist, Data Scientist, or Machine Learning Engineer, it is critical to have data when modeling and testing assumptions. This package is still under development.
pythonpackageWeb Scraping
2018
Trying out Beautiful Soup to scrape and wrangle data from HTML pages.
What is Data Science without the data? Getting data from well-curated sources, like pre-built flat files or an API, still requires processing and data wrangling. However, scraping sites for content can add another level of complexity. What happens if data is missing? What happens if a table moves? What happens if columns change order? How should data types be processed and stored? What happens if the site is down? The Python community has created several packages to help process data scraped from an HTML page. Here, I try out Beautiful Soup.
Stock ETL
2018
ETL pipelines for historical and real-time stock data, with a clear data model and robust error handling.
Stock data is fantastic data to use in practicing data handling, modeling, and prediction. There are two general categories of stock data, historical and real-time. Models may be built using historical data and evaluated using real-time data. The primary purpose of this project is to practice building ETL pipelines that retain a clear data model and handle event errors appropriately. This is an ongoing project and serves as the foundation and data source for downstream modeling and event predictions.
Insight Demo
2015
A proof-of-concept mining masked text-messaging data for sentiment, built at Insight Data Science.
Fellows at Insight Data Science are expected to rapidly build a unique and novel data project for presentation to companies. The goal is to demonstrate the ability to develop, from scratch, product insight from data. The specific project I worked on was fun and challenging and involved mining masked text-messaging data for a type of sentiment. Although the code is very clearly not satisfactory for production, it sufficed in generating a proof of concept and clearly demonstrated strong structure in the data.