If you frequent Hacker News, you’ve probably seen Sam Rose’s work. He’s the mastermind behind “visual, interactive essays” that are educational, beautiful, and also just mesmerizing.
Visit Sam’s site to experience his work
At P99 CONF, he’ll be bringing to life his work on “Fearless Quantization,” a topic that Sam tackled as part of his role at ngrok.
Note: P99 CONF is a technical conference on performance and low-latency engineering. It’s virtual, free, and highly interactive. This year’s agenda spans Rust, Tokio, Zig, C++, Go, eBPF, observability, io_uring, kernel, distributed systems, and of course AI/ML.
We hope you’ll join us live on October 21-22 to hear the talk and chat with Sam. In the meantime, let’s get to know more about him!
How do you answer the dreaded “tell us about yourself” question?
My LinkedIn profile says “I write code and stuff” and it’s said that for about… 15 years. It has stood the test of time.
Recently my career has taken something of a turn. I spent over a decade doing SRE and backend product work, now I write educational content on the web. I have the time to go deep on things I’m curious about, there’s no ticket queue to fight through or pagers going off. It’s lovely.
Computers are endlessly fascinating, and it’s a privilege to be able to get paid to understand them and then help other folks understand them.
What’s the most interesting project that you’re working on right now – or hoping to start soon?
I decided to have a go at porting Kubernetes to TypeScript so that I could run accurate clusters in the browser, you can see the project here: https://github.com/ngrok/webernetes and a post I wrote about probing that makes use of it: https://ngrok.com/blog/probes.
The thinking was that when people show real examples of what clusters do in web content, they’re either spinning a real one up in some backend or they’re just showing pictures and console output. The former takes a long time, can be flakey, costs money, requires ongoing maintenance. The latter is fine but I kinda think we can do better.
So with LLMs now really good at writing lots of code I thought I’d try taking on this ridiculous project. I think it’s really cool! I’m ecstatic with how the probes post turned out, if you check it out I’d love to hear what you think!
What will you be talking about at P99 CONF?
At the start of 2026 I got fascinated by local models. I’d try any new LLM that came out and would fit on my laptop. This meant I bumped up against quantization and was confused by it. It seemed absurd that you could just… make the model 2, 4, even 8 times smaller and it still works.
So I looked into how it worked and it was much simpler than I thought, and the results were undeniably good. I started to wonder how I could do my own quantization and measure the degradation from the original.
So my talk, “Fearless Quantization” walks through that journey and aims to equip you with some tools you may not have known were available to you. Tools that can quantify how much worse a quantized model is compared to the original. I think this is going to become ever more valuable knowledge the better local models become.
What other P99 CONF talks are you most looking forward to – and why?
I’m a sucker for anything that digs into how the kernel works, so Avi Kivity’s talk, Daniel Sedlak’s, and Pavel Emelyanov are all going to be great.
Dan Odell’s talk is extremely relevant to my interests with the sorts of things I build. You might be surprised to learn that webernetes, my Kubernetes port to TypeScript, runs fully on the main thread. I considered not doing that but it never became a problem so I left it. I’m sure Dan will have great ideas for making it better though.
What do you like most about P99 CONF?
I have to be selfish here and talk about my speaker experience, but I’m a big fan of pre-recording! It gives me a lot of space to use the visual elements from my blog posts in my slides, and line up the timing of what I say and what I show to be just right.
Any performance-related blog/article/paper/book/video recommendations for the P99 CONF community?
I maintain a list of my favourite links of all time at https://samwho.dev/links and I’m going to share a great performance related one from there: https://igoro.com/archive/gallery-of-processor-cache-effects/ all about measuring the cost of overflowing into the different levels of CPU cache. A little old at this point but still excellent.
