High Performance Computing & AI
Practical notes on Python, JupyterHub, Kubernetes and AI for science — from the San Diego Supercomputer Center.
Popular categories
Recent posts
-
Keeping interactive tutorials up to date with an orchestrator and AI subagents
How to keep browser-based tutorials working with an orchestrator, a browser executor and a documentation improver, instead of hoping nobody follows the outdated step.
-
Write slides in Markdown with Marp when working with AI agents
Why keeping presentations in Markdown and building them with Marp works well with AI agents, with a clean reusable theme and an example repository.
-
A Marp theme that reproduces the official SDSC slide template
An open source Marp theme that reproduces the official SDSC slide template, extracted from the brand guide PowerPoint file, with a published HTML preview.
-
Measuring the zero-level offset of the GNILC dust intensity map with HI4PI
The CIB monopole was never subtracted from the GNILC dust input of the PySM d9/d10/d11 templates because of a NumPy indexing bug; this notebook measures the real zero-level offset of that map from the HI4PI HI correlation.
-
Managing projects with GitHub and AI agents: the structure problem
In my previous post about managing projects with GitHub and AI agents I described how I organize events and workshops with a single GitHub repository as the source of truth, issues for every activity, and an agent per collaborator.
-
Living with a local 27B LLM on a Mac mini: capable enough and (almost) fast enough
I run an Apple M4 Pro Mac mini in the office as a private LLM server: a 27B Qwen 3.8 model, served over my Tailscale tailnet, used every day by my coding agent to handle emails, Trello cards, and general project work.
-
Mac mini as a private LLM server: the technical setup
This is the technical companion to the experience post: how the Mac mini is set up, how every service runs, and where the configuration lives. The benchmark details (prefill, decode, context stress test, agent context sizes) are here too. The data-center comparison is in the comparison post.
-
Offline speech-to-text on Linux with nerd-dictation
If you want offline speech-to-text on a Wayland desktop, nerd-dictation is a great choice: it records from your microphone, transcribes locally (no cloud), and types the result into the focused window.
-
Same 27B model, two worlds: Mac mini vs a data-center GPU
How much does datacenter hardware buy you over an Apple Silicon dev box? To get a feeling for it, I set the NRP managed LLM service, which hosts Qwen3.8-27B (api name qwen3-small) in bf16 on their Nautilus GPU fleet, against my Mac mini (the high-level story is in the experience post, the full setup in the setup post), which serves the same model quantized to NVFP4 on the MLX engine.
-
SLIC superpixels on the HEALPix grid
A recent healpy issue proposes a slic function that generates superpixels on the HEALPix grid, adapting the SLIC algorithm to the sphere, with a complete working implementation attached. This post is what I found testing it.
-
PySM: produce CAR output at the smoothing stage
PySM 3 only builds HEALPix skies. Here is how to produce CAR pixelization output at the smoothing stage with a single spherical harmonic transform, and a validation against the smooth-then-reproject route.
-
healpy 1.20.1: last release with macOS 14 wheels
The healpy 1.20.1 release includes bug fixes and is the final release to provide macOS 14 arm64 wheels before the GitHub runner retirement.
-
A practical primer on software citation
Software forms the foundation of modern research, yet citing code effectively remains a challenge across scientific disciplines.
-
Gateways 2027: a conference for and by the community, coming to San Diego
At the closing of Gateways 2026 in Washington, DC, we announced that Gateways 2027 will take place in San Diego, hosted by the San Diego Supercomputer Center (SDSC) at UC San Diego.
-
Managing projects with GitHub and AI agents
Organizing conferences and workshops can quickly become overwhelming when plans, decisions, contracts, and meeting notes are scattered across multiple platforms.