Rick Lamers' blog

KV Cache, Made Visible: Qwen3-0.6B on Apple Silicon

Mon, 20 Apr 2026 12:00:00 +0200

A minimal Qwen3-0.6B in pure PyTorch, running on Apple Silicon, with a live UI that makes the memory and latency trade-offs of KV cache physically visible. Toggle the cache off and watch attention go quadratic in real time.

Getting Plymouth Boot Splash Working with NVIDIA 580 on Debian 13 Trixie

Mon, 29 Dec 2025 17:00:00 +0100

A deep dive into making Plymouth boot animations work with NVIDIA's proprietary driver on Debian 13, involving kernel module timing, DRM framebuffer initialization, and initramfs configuration

Gemini 3.0 Spotted in the Wild Through A/B Testing

Thu, 16 Oct 2025 12:00:00 +0200

Testing Google's highly anticipated Gemini 3.0 through AI Studio's A/B feature using SVG generation as a quality proxy

LLMs as general reasoners

Mon, 21 Aug 2023 12:00:00 +0200

Measuring in-context learning performance to unlock further progress

Streaming a USB camera from a Raspberry Pi (4B)

Sun, 20 Aug 2023 16:52:00 +0200

How to stream a USB camera from a Raspberry Pi, tackling issues one by one.

Writing my own ChatGPT Code Interpreter

Wed, 17 May 2023 12:00:00 +0200

It's called GPT-Code UI and is now available on GitHub and PyPI

Writing resources

Mon, 15 May 2023 12:00:00 +0200

Useful writing resources

About

Mon, 01 Jan 0001 00:00:00 +0000

Hi there! Thanks for checking out my blog.

Update: follow Coding with Intelligence (CoWI) for curated updates about LLMs and Machine Learning: codingwithintelligence.substack.com

My name is Rick Lamers. I’m an AI Researcher and I live in Amsterdam.

Previously, I ran a startup called Orchest. We built a workflow orchestration tool not unlike Apache Airflow. We had an open source project and offered a managed cloud service to run said open source project. We raised $4.15M but ended up giving the capital back to our investors (Gradient Ventures, Basis Set Ventures, Seedcamp, Script Capital, Data Community Fund and angels like Anthony Goldbloom) because we believed the market was too commoditized. We were able to return about $2M to investors. In addition, we were able to sell some of the developed IP in a 5-figure deal.