Build locally.
Rent the peak.
Hands-on notes from my own AI experiments: local LLMs, QLoRA fine-tuning, cloud GPUs and the cost decisions behind them. The goal is simple: use enough technology to solve the problem, not more.
Field notes from real experiments
I have deliberately consolidated the site around a few deeper articles rather than publishing lots of thin pages. These are the decisions, measurements and workflows I am actually using.
Runpod for Beginners: What I’d Do Differently on a First GPU Pod
A practical first-run guide based on my own Runpod setup: choosing a GPU, handling storage, validating before paid compute, and avoiding wasted GPU time.
Runpod RTX A6000 48GB for QLoRA Fine-Tuning: My Real Experiment
My actual QLoRA workflow using an RTX A6000 48GB on Runpod, including local baseline results, training configuration, storage decisions and the metrics I want to beat.
Runpod vs Buying a Local GPU: My Actual Cost Decision
Why I kept a modest local AI machine and chose rented NVIDIA GPUs for peak workloads instead of buying a dedicated GPU immediately.
How I Run a 4B LLM Locally and Move Training to Runpod
My hybrid AI workflow: Qwen3.5-4B locally for fast iteration, then Runpod for CUDA and QLoRA training when the local hardware stops being the right tool.
Why I use Runpod
My local machine handles development, inference and validation. I use Runpod when I need NVIDIA CUDA, more VRAM or short bursts of compute that would be poor economics to own.
Try Runpod with my referral link →The nice part is the random credit: Runpod gives each of us $5–$500. It kicks in after your first $10 of spend. I’m actually using Runpod for my own AI experiments, so this is not a theoretical referral page. I got $5 and I’m still putting it to work. Maybe you’ll have better luck than me. Runpod controls the offer and can change the terms.
Prepare locally. Validate locally. Rent the GPU that removes the constraint. Save the artefacts. Shut the compute down.
About Ripper Aussie
An independent personal technology site from Australia focused on practical engineering, AI experimentation and cost-conscious infrastructure. I publish measured results when I have them and label planned experiments as planned rather than pretending they are completed benchmarks.