Dustin Morgan

About

Thu, 22 Oct 2026 00:00:00 +0000

I’m Dustin. I live in Colorado with my family, and when I can get to the mountains I snowboard.

I taught myself computers as a kid. I built my first paid website around 14 and my first Linux server not long after. As a teenager I spent way more time than was wise in the late-’90s FTP and IRC scene, which is where I really learned networks, Unix, and how systems break. I walked away from all of that before college and never went back, but it’s a big part of why I take security seriously.

Now

Thu, 22 Oct 2026 00:00:00 +0000

This is a now page. It’s whatever has my attention at the moment, and I update it when that changes.

Day job

LLM features for Comcast’s internal BI platform, mostly retrieval, multi-hop questions, and keeping the token bill down.

Agent Neo

Agent Neo has gotten a lot better than the public repo. I’ve been training the harness instead of the model, and it went from solving 5 of 12 hard incident investigations to 10 of 12. I’m cleaning it up for a new open-source release. The write-up.

Projects

Thu, 22 Oct 2026 00:00:00 +0000

Agent Neo

Agent Neo is my open-source agent for investigating infrastructure incidents. It’s one Go binary that drives the tools already on your machine (gh, kubectl, aws, ssh, database clients) and can open a PR when it finds the fix. brew install open-auto-intel/tap/neo. The write-up.

Rebooting This Site

Thu, 22 Oct 2026 00:00:00 +0000

The old dustinmorgan.org was a three.js demo with a bio I wrote in one night. This is the replacement.

Training the Harness Instead of the Model

Tue, 20 Oct 2026 00:00:00 +0000

I fine-tuned dozens of models to make Agent Neo better at incidents, and none of them beat a better harness. So I started training the harness. It went from 5 to 10 out of 12 hard investigations.

Coming Soon: CIP and CTP

Mon, 28 Sep 2026 00:00:00 +0000

Two things I’ve been building in private: running big open-weight models across the machines you already have, and training on them too.

What I Got Wrong About SQ4

Wed, 26 Aug 2026 00:00:00 +0000

I published a 4-bit quantization format and called its quality indistinguishable from FP16. Once I measured it properly, standard 4-bit formats beat it. I’ve retired it.

SQ4 Is Live: Whitepaper + llama.cpp PR

Wed, 13 May 2026 00:00:00 +0000

We published the SQ4 whitepaper and opened a PR to bring percentile-calibrated 4-bit quantization to llama.cpp.

Agent Neo: An Incident Agent That Uses Your Existing Tools

Tue, 05 May 2026 00:00:00 +0000

Neo is an open-source Go agent that investigates infrastructure incidents with the CLIs you already have. It once filed 28 bug reports about its own ability to file duplicate bug reports.

v1.5.2: Fused Inference, FP16, and a Tokenizer That OOM'd a 64GB Machine

Fri, 01 May 2026 00:00:00 +0000

Inference +28%. Training +54% at dim≥1024. The tokenizer was O(n²) and nobody noticed until we fed it 48MB.

Benchmarks: I Owe PyTorch a Fair Fight

Mon, 27 Apr 2026 00:00:00 +0000

The 1.54x number has an asterisk. I ran PyTorch with default settings. Here’s what I still owe.

Why I Wrote a GPU Training Engine in Go

Mon, 20 Apr 2026 00:00:00 +0000

I got tired of Python’s toolchain for fine-tuning. So I wrote my own — and accidentally built something faster.