Benchmarking DeepSeek V4 Flash and GLM 5.3 Flash on a MacBook Pro – why parameter counts only predict the start, why the built-in draft model slows things down, and what it costs to keep your RAM back

What active parameters don't tell you about running big MoE models on a Mac

Why the most important skill with AI coding agents isn't prompting or curating context – it's designing how they verify their own work

Feedback loop engineering

How to sandbox coding agents while maintaining development flexibility - the ideal balance between security and convenience

Coding agents in secured VS Code dev containers

How do I mix local LLMs with cloud ones and code gen with continuous learning

My current development tooling setup

Outright rejection of gen AI won't get us the best outcomes

The leftist case for generative AI

I hereby renounce years of suffering with native extension compilations breaking

Using Distroless in production instead of Alpine

How Lisbon has changed in the last 30 years

Lisbon Story 1994–2024

How to set up another drive for a home setup

Expanding TrueNAS Scale with a new drive

Part two of the series Building applications using embeddings vector search and Large Language Models

What are Large Language Models and how to run open ones on your device

Part one of the series Building applications using embeddings vector search and Large Language Models

Understanding embeddings and how to use them for semantic search

Getting up and running, gotchas I came across and my favourite apps

Installing and configuring TrueNAS Scale for home use

Starting from LLMs and capitalism and ending up with experimental fixed focus macro photos on my phone

Delightful rabbit hole #1

How to set up an open source NAS with a second hand tiny PC that outperforms commercial NAS hardware

Build a cheap but fast home NAS system

How to run a low cost OpenVPN server on GCP (or use Tailscale) to access your NAS anywhere even if your home internet is on a shared IP

Safely access your home NAS with an OpenVPN server on GCP

Give app servers a REST – CDNs and APIs are the new LAMP

I was interviewed by Creative Lives in Progress magazine while working at ustwo to show people how it is to work in the creative industry

Creative Lives interview

To make Git (and Github) more useful for collaboration on non-textual content

Why and how Github should have binary diffs

Out with the cars, in with the cargo trikes

Finding an affordable electric family cargo bike in London

Take control back from silos to make the Web personal and open again

Indiewebify yourself

Stop installing development environments locally

Containerise everything

Why YOU — not Henry Ford — should design your working life

The first iteration of a product should be about learning

Build your Minimum Viable Product

What I learnt trying to untangle the async spaghetti

An introduction to jQuery Deferred / Promise and the design pattern in general

CORS can be tricky, especially if you need to do something beyond the basics like authorisation or custom headers

Getting Cross-origin resource sharing with complex jQuery AJAX requests

Including the shocking loss of biodiversity around the World through audio recordings of wildlife

On the fragile relationship between nature conservation and economic growth

Instead of terminal editors with steep learning curve, this guide helps you to tunnel Sublime Text in to your server for remote editing – with sudo

Setting up rmate with Sublime Text for remote file editing over SSH

Python CLI that turns Claude Code transcripts into readable HTML, Markdown, and a browsable TUI

Daniel Demmel

Projects

Claude Code Log

Online LLM Tokenizer

flaunt-it*

Recent Writing

What active parameters don't tell you about running big MoE models on a Mac

Feedback loop engineering

Coding agents in secured VS Code dev containers

My current development tooling setup