Blog
Writing on GPUs, LLMs, MLOps, Kubernetes — and mindset · 3525 posts
#2026-03 765#english 592#culture 264#deep-dive 254#kubernetes 249#career 229#ai 219#llm 209#devops 196#2026-04 146#security 143#database 114#observability 113#communication 109#history 107#architecture 100#productivity 96#finance 88#economy 84#mindset 81#psychology 80#ai-papers 79#food 78#it 78#travel 78#deep-learning 77#japanese 77#networking 77#performance 72#business-travel 70#linux 70#gpu 69#ai-agent 66#cs-fundamentals 63#postgresql 61#rag 58#self-improvement 55#learning 53#mlops 53#python 52
The Palantir Ontology — reading, from the official docs, the operational layer that puts "action" on top of data ♪ Listenable
What Palantir calls an ontology is not a philosophical term but a layer of the product. Six building blocks — objects, properties, links (semantics) and actions, functions (kinetics) — the path from a dataset to an objec
2026-09-11 · 10 min read #palantir#ontology#foundry#aip#data-platformAnatomy of an AI agent harness — 35 files that make the rules enforce themselves
Same model, yet in one repository the accidents repeat and in another they do not. The difference is the harness. A file-by-file dissection of the harness in a repository where an AI accumulated 1,491 commits alongside a
2026-09-11 · 13 min read #ai#agent#harness#claude-code#ci-cdWhat a bastion is in the cloud, and how to use one well on AWS — ProxyJump measured, SSM Session Manager, EC2 Instance Connect Endpoint
A bastion is the host that reduces the doors into a private network to one. Go through a jump host with OpenSSH ProxyJump in a homelab and measure what the target server actually sees, then compare the two things AWS rec
2026-09-10 · 8 min read #aws#bastion#ssh#ssm#securityThree ways to put a proxy cache in front of Docker Hub — registry:2, Nexus, ECR pull-through cache
Read the Docker Hub rate-limit headers yourself, stand up registry:2 as a pull-through cache and pull the same image three times, then sort out where a Nexus proxy/group repository and an ECR pull-through cache each fit,
2026-09-10 · 7 min read #docker#registry#nexus#ecr#devopsCloudNativePG deep dive — who looks after PostgreSQL inside Kubernetes
Why it does not use a StatefulSet, why PID 1 in the pod is not postgres, what the three services do, and why a base backup and WAL must be combined before anything can be restored. The features and role of CNPG, explaine
2026-09-10 · 7 min read #kubernetes#postgresql#cloudnativepg#backup#operatorScanning the production image with Trivy — 768 findings, and the real work was three packages and one Dockerfile line
From a signature-verification bypass in the login library (CVE-2026-27962) to a digest-pinned base image that could never receive OS security fixes. The scanner counting itself, false positives in secret detection, and a
2026-09-10 · 6 min read #security#cve#trivy#docker#pythonVBR MP3 breaks transcript clicks — CBR vs VBR, measured
Clicking a transcript line jumped to the wrong place. The cause was the audio container, not the timestamps. Why the Xing TOC in a VBR file drifts up to 0.9 seconds, why CBR stays within 0.04, and what 96 real episodes s
2026-09-10 · 6 min read #audio#mp3#ffmpeg#web#debuggingHow Reasoning Models Are Built — From Chain-of-Thought to GRPO, in Twelve Papers
o1, R1 and QwQ are not new architectures. They are long output optimised toward a gradeable objective. This traces the lineage through twelve papers — Chain-of-Thought (2022), STaR, PRM, DeepSeek-R1's GRPO, s1's budget f
2026-09-06 · 12 min read #ai#llm#reasoning#reinforcement-learning#grpoA Three-Line Config File Killed Four GPUs for a Week — How containerd Drop-in Merging Really Works ♪ Listenable
The nodes stopped advertising GPUs. The nvidia runtime was written out cleanly in the config file, but it was nowhere to be found in containerd config dump. The culprit was a three-line registry config I had added a week
2026-08-26 · 12 min read #kubernetes#containerd#gpu#nvidia#troubleshootingAI for Everyone, Part 6 — Drawing Digits From Words With a 1.11M Diffusion Model
We built a conditional diffusion model with 1.11 million parameters that draws a 0 when you type "zero". The forward process that adds noise is a single formula; the reverse process that restores the image is that same l
2026-08-24 · 8 min read #ai#diffusion#ddpm#generative#pytorchAI for Everyone, Part 5 — Colourising Photos With a 0.47M U-Net, and Why the Colours Came Out Washed Out
The smallest model in this series — a 472K-parameter U-Net — restored colour to greyscale CIFAR-10 images. Shapes survived intact, but the colours came out noticeably washed out. That is not a capacity problem; it is a c
2026-08-23 · 7 min read #ai#computer-vision#unet#colorization#pytorchAI for Everyone, Part 4 — Captioning Images With 1.37M Parameters, and Why Part 3 Bug Was Absent Here
We joined a CNN encoder to a transformer decoder and captioned Fashion-MNIST images. With 1.37 million parameters and ten minutes of training, the label hit rate reached 91%. This part looks at what cross-attention does
2026-08-22 · 7 min read #ai#captioning#multimodal#transformer#pytorchAI for Everyone, Part 3 — Loss of 0.0017, Accuracy of 7.5%: The Culprit Was One Padding Slot
We built a VQA model — one that answers questions about an image — with 1.48 million parameters. Training loss fell to 0.0017 while accuracy sat at 7.5%, worse than guessing. The cause was not the model but a single line
2026-08-21 · 8 min read #ai#vqa#multimodal#debugging#pytorchBuilding a Hands-On Learning Platform on a Home Server — Putting Kubernetes Inside an Unprivileged Pod
A write-up of putting a learning platform on a 7-node Kubernetes cluster at home: a learner presses a button, a dedicated container comes up, a browser terminal attaches to a real Linux shell, and at every step the serve
2026-08-20 · 16 min read #kubernetes#cilium#security#homelab#kwokLabHub — I opened a hands-on platform where you learn on a real server, in the browser
Some things do not stick from reading. There is no substitute for attaching to a server, typing a command, getting it wrong, and finding out why. LabHub is a learning platform that puts that loop inside the browser. Pres
2026-08-20 · 8 min read #labhub#kubernetes#devops#education#hands-onAI for Everyone, Part 1 — Training a 16M-Parameter Language Model From Scratch in 15 Minutes
Training a language model from scratch on a single GPU. With the TinyStories dataset and a 16-million-parameter decoder-only transformer, we produced readable English fairy tales in 15 minutes. We look at why the causal
2026-08-19 · 9 min read #ai#llm#transformer#pytorch#hands-onFourteen Things Table Tennis Taught Me: Skill Makes the Relationships, a Title Blocks the Growth
Fourteen principles from years of playing table tennis, and how they carry over into work and life. The structure that turns skill into an entry ticket to relationships, why playing only for fun never makes you better, a
2026-08-16 · 11 min read #mindset#growth#deliberate-practice#sports#careerWhat It Means to Be Smart: What Gets Measured and What Does Not ♪ Listenable
The word intelligence has three different things mixed into it. The general factor as a statistical regularity, CHC as the mainstream model of psychometrics, and the social usage we reach for when we size someone up in d
2026-08-16 · 14 min read #mindset#intelligence#psychometrics#expertise#evidenceThe Power of Small Habits — The 21-Day Myth, the 66-Day Median, the 18-to-254-Day Reality
The claim that a habit takes 21 days has no research behind it. That number came from a plastic surgeon's observation in 1960. The most-cited habit-formation study, from Lally and colleagues, actually reported an average
2026-08-16 · 12 min read #mindset#habits#psychology#self-improvement#2026-08How Not to Wait for Motivation — Drive Is Not So Much the Cause of Action as, Often, Its Result
Why does the plan to start once the drive arrives fail structurally? The order of action and mood as shown by a behavioural activation meta-analysis, the autonomy, competence and relatedness of self-determination theory,
2026-08-16 · 12 min read #mindset#motivation#psychology#self-improvement#2026-08