Projects

Developer Matt Coles has built lgtmaybe, a provider-agnostic PR reviewer supporting six model backends with a single --provider flag, shipping as a PyPI CLI and GitHub Action. He also maintains a homelab of Raspberry Pis, NAS, Framework Desktop, DGX Spark, and Dell OptiPlexes running Tailscale for remote access, and fine-tunes open-source vision models like Qwen using Unsloth on local hardware.

Projects A few things I’ve built or am tinkering with - more will land here as they hold up. lgtmaybe lgtmaybe A provider-agnostic PR reviewer. Six model backends behind one --provider flag, and no static keys in your secrets for the cloud ones - Bedrock and Vertex auth keylessly over GitHub OIDC. Ships as a PyPI CLI and a GitHub Action over one core. Python, built ports-first with AI doing most of the typing. I wrote up how I built it /posts/building-lgtmaybe/ , and the source is MattJColes/lgtmaybe https://github.com/MattJColes/lgtmaybe . The homelab the-homelab A few Raspberry Pis, a NAS, a Framework Desktop, a DGX Spark, and a stack of Dell OptiPlexes and other mini PCs. It all runs on Tailscale so I can get to it from anywhere, with a lot of containers going at any time to self-host things and run local AI. Fine-tuning local vision LLMs fine-tuning-local-vision-llms Fine-tuning open-source vision models like Qwen with Unsloth https://github.com/unslothai/unsloth on my DGX Spark and Framework Desktop. Mostly for fun - I want to see how good a model I can run at home is at OCR and reading images, and I like experimenting with swarms of AI agents while I’m at it. More soon more-soon Whatever I’m building next: a Python tool wired up to an AI agent, an agent-orchestration experiment, or some homelab automation. I’ll write them up if I think they’re worth sharing.