Add JSON-LD structured data to rentals, equipment, services and job listings so search engines understand your marketplace pages, with examples for each type.
Continue readingHow.nz - AI Development Blog
Practical notes on ML inference, GPU kernels, AI agents and shipping software, by Lee Penkman, builder of App.nz, Netwrck and Text Generator.
Learn how to stream LLM tokens to the browser with server sent events, cache repeated prompts and measure time to first token in a small Python service.
Continue readingA practical tutorial on how an AI voice call works, from microphone to speech to model to voice, with latency tips and a Python loop you can adapt.
Continue readingThe RTX 5090 is NVIDIA’s Blackwell-architecture consumer flagship – 21,760 CUDA cores, 32GB GDDR7, and 209.5 FP16/BF16 TFLOPS for $1,999 MSRP. It is the most powerful consumer GPU ever released for machine learning workloads. But raw specs don’t tell the whole story: framework compatibility issues, VRAM limitations vs datacenter cards, and wild street pricing all factor into whether it’s the right pick for your ML workflow.
Continue readingYou plug in a shiny new RTX 5090, install PyTorch, run torch.cuda.is_available(), get True, and then your training script crashes with a cryptic CUDA error. The problem: PyTorch’s CUDA availability check only verifies that a CUDA driver exists. It says nothing about whether the installed PyTorch binary actually ships kernels for your GPU’s architecture.
If you read my earlier post on teaching kids programming with Minecraft and Lua, you know I think game-based coding is one of the best ways to get kids hooked on programming. But ComputerCraft isn’t the only option. Let’s compare the four major platforms parents and educators actually use.
Continue readingWhen DeepSeek-R1 dropped in January 2025, it shook the AI world by matching OpenAI o1 on reasoning benchmarks using pure reinforcement learning – and releasing the weights. Over a year later, the reasoning model landscape has shifted dramatically. Let’s see where things stand in March 2026.
Continue readingDictatorFlow is a voice-to-text platform that ships as a single 23MB native binary for Linux, Windows, and macOS. It evolved from VoiceType, a Go-based Linux-only CLI tool, into a full cross-platform application rewritten in Zig with local GPU inference, multi-provider cloud fallback, and a web…
Continue readingProduction ML inference is bottlenecked by memory bandwidth, not compute. Every PyTorch operation launches a separate GPU kernel, allocates intermediate tensors, and round-trips through global memory. CuteDSL fixes this by fusing multiple operations into single Triton/CUDA kernels – delivering up…
Continue readingIf you run Linux services long enough, you eventually pick a side: Supervisor or systemd. The right answer is usually: systemd for the host, Supervisor when you need a lightweight, app-level process manager inside a container or a legacy stack.
This post is a practical comparison and then a walk-through of monitoring strategies, including how we wire AI agents into monitoring to auto-fix common outages (like what we keep in ../netwrck/monitoring).
