Cameron Yule

LLMs for Code Reviews

Thu, 24 Jul 2025 13:46:14 +0100

I’ve been using LLMs for software development for a few months now, mostly from the command-line using tools such as Aider. Lately I’ve been investigating asynchronous LLM workflows, and enabled Gemini Code Assist for GitHub which reviews pull requests for correctness, efficiency, maintainability and security.

It’s still early days, but I’ve been impressed by the quality and usefulness of the review feedback from Gemini, particularly when working as a solo engineer. In a team environment it appears this type of tooling could catch a non-negligible number of issues as a first pass, offering additional quality control on what we’re asking our teammates to review.

Contributing to Nix

Tue, 15 Jul 2025 16:39:32 +0100

I recently made my first contribution to Nixpkgs, the package collection for NixOS, and wanted to document a few notes and learnings from that process. Nix is an ecosystem of tools which together enable reproducible, declarative and reliable systems. It’s comprised of a package manager (Nix), a packages collection (Nixpkgs), a functional language (Nix expressions), and a Linux distribution built using those tools (NixOS).

Nix first caught my attention while I was researching the concept of reproducible builds, which enables verification that binaries were created from a specific version of their source. While reproducible builds offer many benefits, they’re also becoming increasingly important in securing the software supply chain. Nix on its own does not guarantee bit-for-bit reproducibility, but offers the tools to achieve it and reports on the reproducibility of NixOS, which is typically above 99%.

Linting GitHub Actions

Tue, 08 Jul 2025 14:04:47 +0100

I’ve recently started using GitHub Actions – e.g., this website is published automatically to GitHub Pages by an action. While they’re convenient, I’ve also read many critical articles, notably from a security perspective. A few recent examples:

There’s clearly sharp edges we need to be mindful of when working with actions, so in addition to reading these types of articles, I also looked for tooling which could help. This led me to configure actionlint and pinact for my projects which rely on actions.

LLM Structured Output

Fri, 04 Jul 2025 15:15:57 +0100

While integrating an LLM into a Python application, I realised that their default response format of unstructured text wasn’t always practical. This led me to discover that most current models and tooling support requesting structured output, such as JSON. For example, see Google Gemini structured output support.

The tool I’m using for LLM integration — the Python API of Simon Willison’s excellent llm library — has support for structured output via schemas, however the llm-mlx plugin I was using for local model access did not. (MLX is an Apple framework for running models on Apple Silicon, typically giving higher performance.)

About Me

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

Testing creating a static page (About.md)