garage

Things I’ve built along the way, some for work and most because I wanted to find out how they worked.

running in the wild

Used in production or benchmarked against the tools people actually use.

poiesis2025
Runs GA4GH TES tasks on Kubernetes, one Pod per task, with a production Helm chart that an outside organisation has forked and now runs in production. Read the write-up
metis2025
A workflow engine in Rust. Nextflow, Snakemake and CWL plug in behind one trait, and the API talks to its workers over NATS JetStream. Held about 2,000 submissions a second with 10k+ workflows in flight, on one node.
s3z2026
An S3 client written from scratch in Rust, with Python and Node bindings. Downloads up to 3× faster than aws-cli on 17 MB of memory to its 400. The benchmarks are in the repo, so you don't have to take my word for it. Read the write-up
poiesis-mcp2025
An MCP server, so AI assistants like Claude and Gemini can submit and babysit compute jobs on any GA4GH TES server.

Smaller tools and experiments.

poiesisd2026
poiesis without the Kubernetes: one docker compose up gets you a task service with a web UI.
jupyter_s3fs2024
Mounts S3 into Jupyter over FUSE, so remote workflow outputs look like local files.
vpn2024
One command starts an EC2 VPN and connects to it, and the same command shuts it down again, so it only costs money while I'm actually using it. Read the write-up
CampusHub2023
A marketplace for buying and selling on campus, for students by students.
QuickLink2023
My take on the classic URL shortener: custom slugs and a small REST API in Go.