SillyML
6 subscribers
3 links
Download Telegram
Channel created
High velocity ML development with Nix and AWS

Quick start cloud ML development environment with Nix

Read full article → https://mlship.dev/blog/nix-for-ml
Sorry for constant ping, had issues with my autoposting CI
2
nixlab: Own Your Compute

A NixOS homelab template for a self-hosted k3s cluster, built out of frustration with minimal setups and a belief that owning your compute matters

I started this after watching a Dreams of Autonomy video in 2024, bought Beelink mini PCs, fought through Kazakhstani customs, and ended up deeply unsatisfied with minimal NixOS homelab setups that only bootstrapped the cluster but left service deployment out. So I built my own: a fully declarative k3s cluster where Colmena handles the system layer and a NixOS activation script applies Helm charts on top. The entire stack (MetalLB, Longhorn, Pi-hole, WireGuard, Nextcloud, cert-manager) lives in a single vars.nix. Every boot wipes / via btrfs rollback so nodes are always in a known-good state. The deeper reason: I think owning your compute is becoming essential as corporations buy more and more for datacenters.

Read full article →
2
Optimizing the Pilot: My Terminal-First AI Agent Workspace

While everyone builds visual dashboards for AI agent orchestration, I went the opposite direction. Here is why a terminal, home row mods, and Zellij is faster than any GUI for managing multiple AI agent sessions.

Everyone is talking about how to orchestrate AI agents. Nobody is talking about the actual UX of driving them fast. The industry default is visual dashboards, reasoning panels, and orchestration UIs. I went the opposite direction and my workspace is a terminal. This post covers the two tools that made that setup genuinely faster than anything I have tried with a GUI.

Read full article →
1