About the lab · Experimentos

A home lab that publishes its evidence

I'm Vlad Jagodnik, based in Auckland, New Zealand. Experimentos is where I test local AI on the machines I have: language models, quantization, speech, image generation, coding agents and policies that learn from pixels.

I keep the part between a model fitting and a useful system working visible. A result can end in a deployment, a rollback, an incident, or a better question.

Every finding needs its context

State the workload, precision, hardware and units.
Keep failed runs and rejected candidates visible.
Separate warm throughput from first-request latency, and automatic metrics from human quality.
Attach the source revision, qualification limits and downloadable measurement data.

The original website was crafted in ChatGPT and lives in the experiment repository. This publication preserves its experiment catalogue, benchmark explorer and Lab Notes design. The content comes from reviewed experiment records. The editorial inspiration is The Kaitchup — AI on a Budget; the stories and measurements are my own.

The reviewed archive

69 experiment stories, 86 supporting notebooks and 56 figures are available here. GitHub source links require access to the private archive; the readable reports and aggregate exports are public.

Private recordings, raw conversations, credentials, workstation paths and service addresses are excluded. New source folders do not publish automatically: every addition passes editorial review and the public-file privacy gate.