Picking up where part 1 left off. Streaming at the right granularity, subgraphs as Runnables, and the Send API for the moment you realise you need to fan out to N workers and N is not known until the LLM tells you.

I have been building agents with LangGraph for a while and finally sat down to explain the parts that took me longest to internalise. Part 1 covers everything up to and including human-in-the-loop.

I use Obsidian for bullet journaling but missed the immediacy of Things3 Quick Entry. Obsidian Headless unlocked a simple way to add that capture flow back.

View the repo on GitHub Install Loupe from npm When you are building with LLMs there is a very annoying tooling gap. At one end you have console.log() and whatever scraps of logging you remembered to add at 1am. At the other end you have full observability stacks built for hosted apps, retained data, team dashboards and production telemetry. Most of the time, when I am iterating on an agent locally, I do not want either of those things.

The problem of SMS Phishing (AKA Smishing) Email spam has been a solved problem for at least 10 years. As any user of Gmail will tell you, the only time they see spam emails is the occasional time they curiously dip into their spam folder and chuckle at the all-to-obvious clichés used by scammers. However, there is another type of spam that is on the rise and seems to be poorly addressed by Big Tech until now.

This Blog is built with the wonderful Hugo static site generator. It is hosted on GitHub pages too and uses the Soho theme - that’s a lot of great stuff for free. GitHub pages hosts a static site straight off your repo, either via a specific docs folder on master branch or a dedicated gh-pages branch. To build the static directory Hugo just has a simple hugo command to run. Up till now I would run that hugo command after making and changes and then commit and push the changes to my blog repo, simple right?

I want to share with you my first experience working with GitHub Actions. They’re really neat and definitely worth your time if you’re a fan of automation. Background This weekend I was working on a small personal project: a GitHub PR Comment resource for Concourse CI. That isn’t what this post is about really but first a little context. The project I was building is a Go project. When built, it consists of 2 built binaries check and in.

I’ve recently been playing around with Deno - the “secure JavaScript/TypeScript runtime built with V8, Rust, and Tokio”. The reason being is that this lies at the intersection of a couple of my main interests: JavaScript and Rust. I’ve been writing JS professionally now for around 5 years and Rust very unprofessionally for just over a year. Deno was created by Ryan Dahl, the also-creator of Node.js. Ryan introduced Deno to the JS world in a talk titled 10 things I regret about node.

Rust is an awesome language and platform to use, however there’s so much great software already written in c/c++. Luckily it’s not too complicated to make use of c/c++ projects in Rust. In this short post I’ll show you how. From a high-level perspective you can take any c/c++ project, for this example I’m going to use Duktape, the lightweight embeddable JavaScript engine. I’m choosing Duktape because it’s very simple to build it - it’s just 1 .

I’ve spent the last couple of weeks learning about the art and science of Ray Tracing. Ray tracing, for those who aren’t familiar is one technique for generating 3d computer graphics. Ray tracing isn’t the fastest way to generate 3d images but it’s appeal lies in both the realistic effects that can be achieved and in the elegance of simplicity of the technique. This technique is used in movies and for photo-realistic architectural renderings.