Vicki Boykis+

😭

😭

Blog posts we used to have: “how I reduced latency across several hundred machin

Blog posts we used to have: “how I reduced latency across several hundred machines by 200 ms by removing two O(n)^2 hot paths” Blog posts we have now: “Here’s how to get the most out of /ultrathink in the latest release of Claudex 3.0.5”

Curious about people’s ML dependency management strategies in the face of supply

Curious about people’s ML dependency management strategies in the face of supply chain risks. Often pinning dependencies particularly in the torch/HuggingFace ecosystem is the only way to create reproducible training and inference code but that means dependabot isn’t happy.

veekaybee starred jmaczan/tiny-vllm

There are millions of anti patterns that make stuff slower and less elegant but

There are millions of anti patterns that make stuff slower and less elegant but probably the worst offender is adding a network call

There are like only three design patterns you need to use 80% to make stuff good

There are like only three design patterns you need to use 80% to make stuff good and fast and elegant Make it a lookup (hashmap) Split it up into smaller pieces (sharding) Combine it into one large piece (map reduce)

Eye bleach: an actual architecture and engineering blog post lobste.rs/s/ko1ji

Eye bleach: an actual architecture and engineering blog post lobste.rs/s/ko1ji1/lob...

❤️❤️

❤️❤️

I have but I’m looking for something more technical and less biased than both A

I have but I’m looking for something more technical and less biased than both Andrew’s and Jared’s perspectives (both valid, but they’re at the center of the storm.) I’m looking to understand the core of the technical issues

Looking to read an in-depth examination of the project history and code evolutio

Looking to read an in-depth examination of the project history and code evolution of Bun in Zig, along with a technical evaluation of its current Rust re-implementation. I want to check out links to PRs, sidebars on Zig and Rust, the works. Who’s doing this?

veekaybee starred scop/bash-completion

veekaybee starred tcpipuk/llama-go

veekaybee starred QwenLM/Qwen3

Not sure I know what either of those is but the basic idea is that intuition for

Not sure I know what either of those is but the basic idea is that intuition for writing good code is a skill that can be learned over time

Writing good software should be a slow and deliberate craft

Writing good software should be a slow and deliberate craft

Recommended reading if you feel you’ve lost your way , or if you’re having more

Recommended reading if you feel you’ve lost your way , or if you’re having more fun than ever

traces are just logs right

traces are just logs right

and of course I'm so behind 😭

and of course I'm so behind 😭

Everyone working on recsys/embeddings/information retrieval/search is sleeping o

Everyone working on recsys/embeddings/information retrieval/search is sleeping on this newsletter if you're not subscribed. I can't believe I didn't mention it before but it just summarizes the best content in the field every week recsys.substack.com

like, discuss.python.org/t/using-exce...

like, discuss.python.org/t/using-exce...

the actual most revolutionary thing the ai labs have actually done is make it lo

the actual most revolutionary thing the ai labs have actually done is make it look like it's insanely easy to duct tape together python microservices with all their concurrency issues and trained models with all their transformers issues and have them run as one api call

This is the only thing noam shazeer has blogged about publicly and I use it all

This is the only thing noam shazeer has blogged about publicly and I use it all the time medium.com/@NoamShazeer...

Nah, should still work

Nah, should still work

new post: how I develop recently using local models. the tooling is now good eno

new post: how I develop recently using local models. the tooling is now good enough to do agentic workflows and everyone should give them a try! https://vickiboykis.com/2026/06/15/running-local-models-is-good-now/

Were you doing context pooling or a sliding window? lm studio (and likely llama

Were you doing context pooling or a sliding window? lm studio (and likely llama cpp) has options to change how the context window changes

I’ve been able to so far bump the context window to 75% of max but ill probably

I’ve been able to so far bump the context window to 75% of max but ill probably run into the same issue and will want to check out stuff like this arxiv.org/abs/2606.06492

You can try a lower quant and see if that helps any, though the results might no

You can try a lower quant and see if that helps any, though the results might not be as good

It can do whatever you want it to! It does YOLO out of the box which is why I li

It can do whatever you want it to! It does YOLO out of the box which is why I limited it but you can write extensions that control its behavior

👀🤖

👀🤖

LMK if you have questions!

LMK if you have questions!

new post: how I develop recently using local models. the tooling is now good eno

new post: how I develop recently using local models. the tooling is now good enough to do agentic workflows and everyone should give them a try! vickiboykis.com/2026/06/15/r...

The machine learning circle of life

The machine learning circle of life

The good thing is that you don’t need it! Unless you want something fun and shin

The good thing is that you don’t need it! Unless you want something fun and shiny. My laptop is 4 years old

RAM is very sad

RAM is very sad

this is so cool, continuing my local experiments and using pi and lm studio and

this is so cool, continuing my local experiments and using pi and lm studio and gemma to introspect how i use agents locally. On the left, you can literally see the streaming tokens and the context window filling up.

I’m seeing the opposite

I’m seeing the opposite

every day we get closer to looping back around to small purpose-built specialize

every day we get closer to looping back around to small purpose-built specialized models

Oh sorry this pi: pi.dev

Oh sorry this pi: pi.dev

New post: Tagging my blog with BERTopic and LLMs. I suspect as token costs incre

New post: Tagging my blog with BERTopic and LLMs. I suspect as token costs increase, we'll see more blended traditional ML/LLM systems. And, in general, they work really well! https://vickiboykis.com/2026/05/18/tagging-my-blog-posts-with-bertopic-and-llms/

Last week, I gave a keynote at https://appliedml.us/2026/ on machine learning an

Last week, I gave a keynote at https://appliedml.us/2026/ on machine learning and engineering. Like a lot of us, I've had questions and anxiety around software development today. The talk covers why good engineering is still important (and 🌷). https://vickiboykis.com/2026/04/20/build-yourself-flowers/

New post: if you're a developer and you're feeling down these days, it's ok. One

New post: if you're a developer and you're feeling down these days, it's ok. One way to recover is to build stuff for yourself, with the process and tools that you dictate. https://vickiboykis.com/2026/03/04/antidote/

New post: Querying 3 billion vectors https://vickiboykis.com/2026/02/21/querying

New post: Querying 3 billion vectors https://vickiboykis.com/2026/02/21/querying-3-billion-vectors/

My 2025 in review https://vickiboykis.com/2025/12/22/2025-in-review/ #machinelea

My 2025 in review https://vickiboykis.com/2025/12/22/2025-in-review/ #machinelearning

every time i try to write a post that's like "you should do this because it's be

every time i try to write a post that's like "you should do this because it's best practices and good engineering is important", it comes out forced and unnatural until i turn it into "here's what i learned about x, y, z, come on a trip with me." every single time.

A lot of good programming is working within constraints that you only figure out

A lot of good programming is working within constraints that you only figure out after years of deploying to prod. You could name that variable “x” but it’s much nicer to call it “items_sorted_by_price,” especially at 3 am on PagerDuty, and even nicer never to have to declare it at all.

New post on whether mastery in writing software still matters. I decided it does

New post on whether mastery in writing software still matters. I decided it does to me. https://vickiboykis.com/2025/10/20/i-want-to-see-the-claw/

I did a deep dive on current embedding sizes and architectures https://vickiboyk

I did a deep dive on current embedding sizes and architectures https://vickiboykis.com/2025/09/01/how-big-are-our-embeddings-now-and-why/

Thinking a lot about code optimization. Anyone have good blog posts around what

Thinking a lot about code optimization. Anyone have good blog posts around what people mean concretely when they say premature optimization, with code examples?

Locked out of my Twitter account by someone who got into it maliciously. Have fi

Locked out of my Twitter account by someone who got into it maliciously. Have filed a support case but not expecting a lot in terms of platform support since they're short-staffed (although if you know someone still over there...). If you see any suspicious tweets coming from it, it wasn't me. 🫠 This was the phishing email I got, which had links to my actual Twitter settings. If you get something

In the hyperbolic discussions around AI, it's easy to forget that you can get it

In the hyperbolic discussions around AI, it's easy to forget that you can get it to do extremely useful, constrained tasks in development workflows. I wrote post about how one of my favorite smol AI tools, Full line code completion in JetBrains products, works. https://vickiboykis.com/2025/07/16/my-favorite-use-case-for-ai-is-writing-logs/

Had a shower thought today that most programming problems are either search or c

Had a shower thought today that most programming problems are either search or compression. Cache invalidation is search and naming things is compression.

It's been feeling like the local LLM space has really slowed over the past 6 mon

It's been feeling like the local LLM space has really slowed over the past 6 months, curious if it's just me or others have felt it too. If you're using local models, which ones and for what?

What do you feel is missing in educational texts/ blog posts about machine learn

What do you feel is missing in educational texts/ blog posts about machine learning in industry as it exists today?

Just rereading some recsys papers from 2022, could be blanket-applied to LLM eva

Just rereading some recsys papers from 2022, could be blanket-applied to LLM evaluation today https://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/21745

I've come across this naming issue in software a bunch recently: I have a machin

I've come across this naming issue in software a bunch recently: I have a machine learning model, an object model (Pydantic for example), and an MVC model. How do you disambiguate between these different taxonomies in your code?

Wrote a post digging deep into FastAPI path operations #python https://vickiboyk

Wrote a post digging deep into FastAPI path operations #python https://vickiboykis.com/2025/01/14/how-fastapi-path-operations-work/

It’s hopping over here https://bsky.app/profile/vickiboykis.com 🐇🦋

It’s hopping over here https://bsky.app/profile/vickiboykis.com 🐇🦋

What are embeddings?

A very deep dive into deep learning

Where do we go from here

Ended friendship with old internet, new internet not here yet

The art of the long goodbye

What happens when we leave but never truly leave

What we talk about when we talk about The Algo

Algorithms are complex systems that are made up of half code, half humans

Duo, the Push, and the Bandits

Why am I terrified of a 2D owl?

With the Feed as my witness

How to make sense of a senseless world

Scaling Normcore way down

Goodbye on the Mersey, James Tissot, 1881 Hello friends! As you might have guessed from my four-month absence, I can no longer continue writing Normcore on a regular basis.

The luxury of length

Hi Friends,

The last miles and miles of the vaccine

One web dev versus the System

Diamond hands, when memes became power ✋💎🤚

Our new, growing shared and unknowable Internet culture

The blank slate in your hands

How do we rebuild the internet from scratch?

Containerized

Do you like boxes in boxes?

A winter's tale for the end of the year

Normcore 2020 in review

My Hero's Journey to Peloton

First they laugh at you, then they want to buy you

The rules of the game

How do you become a Senior Person in Tech?

Ad block and you don't stop

Who will win, a multibillion dollar industry or a couple of txt files?

The great A/B/C/D divide

What happens to society when we don't stand on the same ground?

You're a sky full of stars

Emerging from the underground

The rise of the non-expert expert

From ones and zeros to millions of frameworks

The satisfaction of the hand-off

I'm not going to go build tables or something, but still...