Here are a couple of projects I’ve created or contributed to.

intervals-search

Natural-language search for intervals.icu workout data using OpenRouter's LLM API. Ask "show me hard bike rides from last month" and get filtered results.

Tech stack: FastAPI, React, OpenRouter (gpt-5-mini), intervals.icu API

Why I built it: Wanted to explore LLM integration patterns and MCP. Also served as a reminder that I prefer backend development - I had Claude write all the CSS.

Code

Homelab

Mini PC running NixOS with Caddy reverse proxy, DNS server, Immich, Miniflux, Calibre, and other services. Plus a Raspberry Pi running Home Assistant OS. My NixOS configuration is highly modular and shared across multiple devices - the home server, my laptop, and formerly my Steam Deck.

Key features:

  • Declarative management under version control for full reproducibility
  • Secrets (DB passwords, API keys) managed centrally with agenix
  • Services accessible only over a Tailscale VPN, with Caddy providing HTTPS
  • Automated backups via Restic

Why it matters: When I break something during a rebuild (which is rare), Nix makes rollback trivial. Outside of my own mistakes, uptime has been 100%.

Personal website

This site is built with Zola and the Terminimal theme, featuring automated deployment to AWS S3. Key technical aspects:

  • CI/CD Pipeline: GitHub Actions workflow that automatically builds and deploys on pushes to main
  • Infrastructure:
    • S3 bucket configured for static website hosting with public access
    • Route 53 for DNS management
    • ACM for SSL certificate management
    • IAM roles and policies configured following principle of least privilege
    • Cloudfront for content delivery and HTTPS
  • Development:
    • Content written in Markdown
    • Local development using Nix flake for reproducible dev environment and consistent Zola versions
    • Git submodules for theme management
    • Favicon generated using DALL-E 3

Building this site didn't require any programming, but it was a chance for me to practice my DevOps skills. While the source code doesn't contain anything too interesting, it's publicly available on GitHub.

gpm-to-spotify

When Google announced they were shutting down Google Play Music, I realized that I would have to find another music streaming service. Spotify was the only other one with a Linux desktop client, so the choice was easy. But there was no way for me to transfer my library from GPM to Spotify without paying some random third party. I figured that if both services had APIs, how hard could it be to do it myself? (Famous last words…) And thus, gpm-to-spotify was born.

This is probably the ugliest, hackiest project I’ve worked on that’s actually worked. It started as a bash script, but then I realized GPM’s (third-party) API only had Python bindings, so that part had to be spun out into Python. Then, on the Spotify side, I realized as I worked on the project that it would need more API calls and error handling than I’d initially accounted for. Rather than refactor my code, I just indented my code and debugged errors as they came up. (Coding horror!)

Mistakes were made, and lessons were learned! Granted, this was far from a production app - I was creating this solely for myself, the GPM side was read-only, and since I had no existing data in Spotify I could just remove everything when I messed up. But if I were to start again, I’d do this very differently. I’d write the whole thing as a Python script and split out the script’s functionality into smaller functions, making the code much more readable and debuggable.

Imperative language interpreter

For a class on programming languages (CSDS 345), I worked in a small group to create an interpreter for a simple imperative language in Scheme. It supports functions (including nested function definitions), break and continue from loops, and try/catch/finally blocks, and has limited support for objects. The interpreter is purely functional, with program state passed as an argument between functions; tail recursion keeps stack size in check. Because it was a class project, I’m not able to post the source code publicly, but it is available upon request.

HTTP web server

As part of a networks class (CSDS 425), I built a basic HTTP/1.1 web server in Java with support for persistent connections. Because it was a class project, I’m not able to post the source code publicly, but it is available upon request.

Simple traceroute

For the same networks class, I built a simple traceroute tool in Python to measure the number of hops to a remote host. Like the HTTP server, I can’t post the source code publicly, but it is available upon request.