Thoughts on AI.

blog / projects / about /

Blog

Containing Chaos: Building A Safe Agentic Pipeline

How to build a secure multi-agent AI pipeline using the Plan-then-Execute pattern, human approval gates, and the principle of least privilege.
BreakLeg: Securely design your AI agent setup

BreakLeg visualizes multi-agent pipelines and checks them against the Lethal Trifecta in real time.
Sandboxed AI Agents and the File Pipeline

Two sandboxed OpenClaw agents, connected by a file.
Changes to reading mode

A simpler design but more difficult training by fewer answer choices in reading mode.
Interval quality and quantity in the Note Reading Trainer

From now on, interval quality and quantity reading skills can be trained.
Reading intervals in the Note Reading Trainer

You can now train reading intervals in the Note Reading Trainer.
Score graphs in the Note Reading Trainer

A graph now shows the scores of the current session.
Note Reading Trainer v0.0.2

Release of a new version of the Note Reading Trainer, now with even more notes!
Implementing Tachyons in a Nuxt.js project

A short tutorial on using Tachyons with Nuxt.
Note Reading Trainer

Release of the first version of the Note Reading Trainer.

© 2026 • amathia

This site uses Umami for anonymous pageview analytics — no cookies, no personal data collected.

Press Esc or click anywhere to close