I’m continuing to put my RTX 5090 through its paces. A few months ago I tried running Qwen3-Coder-30B on it and hooking it into VS Code and Copilot. This unfortunately produced limited results. The code it generated was alright, but it was very inconsistent with tool use even after some troubleshooting. It would just spit the code into the chat window rather than actually updating the files, and sometimes couldn’t even read files. Apparently the root cause of all this is that the Quen model was not configured to produce tool output in a way compatible with Copilot.

Recently xAI released Grok Build, their version of an AI coding agent like Claude Code. I’ve had a subscription to Grok for a while for chat so I decided to give it a try.

After setting up my new graphics card, I began to make use of the 32GB of VRAM to test out some larger models. I started by testing a large image model to generate some images. I already had ComfyUI set up from my previous testing, so I just swapped in a bigger model. The results looked significantly better than what I was able to achieve before, with about the same processing speed. The detail of prompts definitely still matter though and I don’t quite have the knack for it myself. But I found asking Grok to expand what I write into a prompt worked pretty well. Here’s the same prompt of wizard bartenders with v1-5-pruned-emaonly (4GB) and Jib Mix Realistic XL (13 GB).

Despite my better judgement, I just bought an RTX 5090 for about $3,200 with tax. This is by far the most expensive graphics card I’ve purchased since I’ve always stayed away from buying high end cards. You get heavy diminishing returns on price on the high end of hardware, but I thought it would be worth it this time. It also seemed like a good time to buy since RAM prices were spiking due to AI data centers scooping up the supply, but the price of GPUs hadn’t really increased yet.

I’m the kind of person that likes to run tech into the ground. The average lifespan of one of my phones or computers is about 5 years. It’s probably a bad habit of mine to stubbornly no upgrade until there’s a very compelling reason, but I tell myself I’m being frugal. Well I finally hit that point with my old HP laptop I bought in 2019. But my main usage for a laptop is just to have something to do work on and some light gaming while on the go. Something lightweight with enough battery life to get me through watching movies on plane flight. So there was really nothing wrong with the specs on the old machine, but it’s been losing battery capacity rapidly. It got to the point where it would only hold like 2 hours of charge which was barely enough to watch a movie on a plane while not plugged in.

Time for yet another rewriting of this site. The Wordpress structure I was using was so stripped down that it could be statically hosted on S3 so I figured the next logical step is to use a static site generator instead. I decided to go with Hugo which is a Go based static site generator. As a result, I migrated the existing posts, reorganized the layout of the site and added some backdated posts about some projects I did over the last few years. The best part is since the site is static I can keep my existing S3 static site setup but now have a much more lightweight way of updating it than having to spin up a Wordpress server.

My yield curve site yields.paulsulli.com has continued to be convenient this year with all the interest rate hikes we’ve seen. I check it myself and have a few friends that use it too. But one day a month or so ago one of them complained that the site wasn’t loading. That’s weird, I didn’t change anything. Well it turned out that the US Treasury, which I’m using as the source of the data, did make an update. In addition to changing their formatting slightly, they seem to have introduced a ~10 second delay in downloading files. I suspect this is some sort of rate limiting. The problem is that when I slapped this together I simply had code in an AWS Lambda that would run every time the site is loaded and query the Treasury to get data for this year and the previous two years. Because of the new rate limiting it now takes ~30 seconds for the page to load. Not good!

I recently got a Raspberry Pi 4 and started messing around with it. It’s pretty impressive what you can run on such a small machine, but the main reason I got it was to install Pi-hole. Pi-hole is a DNS-based adblock that will work for any device on your network. I already have adblock on my computers, but it seemed like a nice way to block things on my phone and possibly my smart TV. It’s up and running now and it’s helpful for the phone. Unfortunately, it does not black YouTube ads since those are not a 3rd party but hosted by YouTube itsefl. I’ll probably do some more experimenting with the Pi but this is good for now.

Nothing much has changed with what you see with this site, but I’ve made some major changes in the background. Rather than using a Wordpress server on an EC2 instance, I’ve set up static hosting in an S3 bucket and dumped the content there. While I was at it, I also moved my domain from Godaddy to AWS and set up some certificates with a CloudFront distribution. So now you can view the site through S3 with SSL. Neat. And since I could, I also made a subdomain for my yield curve site so all you need to do it go to yields.paulsulli.com rather than holding on to an unsightly S3 link.

Until recently, I’ve been running this site on a managed WordPress host. It was pretty simple for me to set up when I was getting started, but now that I know more about AWS I decided it would be interesting to try moving to an AWS hosted instance. The yearly subscription I have with the other hosting company is about to renew, so it was a good time to move this over.

I’ve been settling into my new job nicely. A big part of it has been learning in depth about AWS in preparation for taking the AWS Certified Solutions Architect Associate exam. I’ve been doing some experimenting as part of that.

It’s been a while since I’ve posted. That’s because I’ve started a new job as a Security Incident Response Engineer at Marcus by Goldman Sachs. Before I left Bitglass I wrote one last blog post which you can find here:

Here is my latest blog post for Bitglass where I discuss the bit Collection 1 breach from January. It was the largest single breach by number of records, though there’s some nuance in that number that I explain in the article. https://www.bitglass.com/blog/collection-1-data-breach

I’m a bit delayed in posting this because of the Christmas break, but here is my latest post for Bitglass about the impact of SaaS downtime on customers.

Here is another post I wrote for Bitglass, this time discussing how unsecured databases and S3 buckets are large a risk for data leakage. Though the post itself features the archetypal hacker in a hoodie, I’m using it satirically. How does he keep getting away with it???

I’ve just finished a new video game related project. EVE Online is a massive multiplayer online (MMO) game where people build space ships and fight each other with them. The fighting part of the game can be exciting and the fact that you can trade game time for currency in the game leads to some impressive headlines about the value of destroyed ships, but a large part of building ships is fairly mundane. I’m talking about mining. Everything in the game is created by players, which means that for every ship destroyed, someone, somewhere, has to mine the materials and build a new one. In addition to other things, I do a fair bit of mining. It’s possible to get enough money in the game to trade for subscription time and keep your character subscribed indefinitely. Unlike other online games, using multiple characters is allowed in EVE. This scales great with mining, so I’ve built up quite the mining fleet over time.

In this post I’ll be showing you the result of my analysis of some WoW private server auction house data. If you don’t know what that is, check out my previous post about it.

I’m almost done with my second post about WoW auction house data, but in the mean time, here’s another article I wrote for Bitglass about what I’m working on there.

I’ve been sitting on some data I collected a while ago, and it’s finally time I do some analysis of it.

For my Master’s degree, I experimented with data visualization in VR, conducted a user study and then wrote a paper about my findings. You can read the abstract and download the full paper here.

After some vacillating, I finally decided to make this site. I’ve been working on a few projects unrelated to my current work and figured I should make a place online to host them. On this blog I’ll describe what I’ve been doing with links to the code and possibly also discuss other technical things I find interesting.