My 2024 in Code: Personal Projects & AI Tools
How AI helped me build useful software while leading PolicyEngine
As 2024 drew to a close, I reflected on my coding journey. Leading PolicyEngine meant less time programming, but I still maintained 3,721 GitHub contributions last year (down from ~4,500 in prior years), spanning collaborations, third-party contributions, and personal projects.
While most of those contributions went to PolicyEngine's core software, where I continue to review code and build with our growing team (see our 2024 year-in-review posts for the US and UK), 2024 was also my most enjoyable and productive year of coding, thanks largely to AI tools like Claude. Here are some of the projects I built alongside our main work:
CodeStitch
codestitch.dev stitches GitHub repos, PRs, and issues into single text files to make it easier to give code as context to AI assistants. It started as a Streamlit app and evolved into a React application in December. If you've ever tried to get AI help with a codebase, you know the pain of context limitations—CodeStitch helps solve that.
PR Improver
pr-improver.streamlit.app suggests improvements to pull requests using LLMs and the GitHub API, comparing against contributor guidelines in the repo. While it currently only works with PolicyEngine repos (since each run costs us money), I'm planning to add support for other repos where you can bring your own LLM API key.
OEWS Explorer
oews-explorer.streamlit.app helps explore the Bureau of Labor Statistics Occupational Employment and Wage Statistics by occupation and geography. I built it to support PolicyEngine's NSF grant application, but it's also useful for compensation research.
Personal Website
I refreshed maxghenis.com with React, creating a cleaner, more maintainable platform for sharing my work and writing.
Looking Forward
AI has transformed programming more than any field I've encountered. You can now ideate, prototype, and build with unprecedented speed and creativity. Tools like Claude and Cursor don't just help you debug—they're collaborative partners in the development process, helping you explore architectural decisions, discover edge cases, and learn new frameworks.
Whether you're a seasoned developer or curious leader, I highly recommend making time for coding in 2025. You might surprise yourself with what you build. I'm excited to share more projects in the coming year!
Want to explore these tools? Check out the links above and let me know what you think—they’re all open source. And if you're interested in contributing to PolicyEngine, we're always looking for collaborators—drop by our GitHub repos.
Others metrics I would test are employment rates, wage growth, and consumer confidence as all of these directly impact economic well-being. Comparing predictions across different AI models could help gauge consensus or divergence in AI-driven forecasts. The challenge is ensuring these models are used responsibly and with an understanding of their limitations. Would love to hear more about how you validated the accuracy of past AI-generated forecasts.