AI experiments, in the open

Working code, design notes and what each experiment cost to run.

• A1

  Ask Abovo

  In progress

  Retrieval‑augmented assistant over this site’s case studies, CV and posts. Serverless proxy, rate‑limited, answers cite their sources.

• A2

  claude-lab

  In progress

  Six sessions from a first Messages API call to multi‑agent orchestration, in Python, behind a MeteredClient that enforces a budget.

• A3

  Accesible por diseño

  Research

  WCAG 2.2 benchmark of public‑sector portals following WCAG‑EM, using axe‑core, Playwright and a multi‑agent review pipeline.