About this notebook

I'm Hasan Asyari Arief. By day I'm a Senior Researcher atNORCE Research ASin Bergen, Norway, building machine learning for the physical world: leak detection from fiber-optic sensing in pipelines, computer vision for subsea inspection, physics-informed networks for energy systems. A growing share of that work is LLM tooling, models that call tools from a constrained catalog and return outputs you can audit.

This notebook is the other half of the job. Working with coding agents and open-weight models every day produces a steady residue of findings that never fit a paper: which practices survive contact with production, what a model actually costs to run, which tool earned its place and which one got deleted. I was writing these notes anyway. Publishing them forces the discipline of checking every claim, and gives me something to point at when someone asks how I work.

A note on who reads this: a surprising share of traffic in this field is AI research agents, not people. I treat them as first-class readers. Every entry has a raw-markdown twin (append .md to any URL) and the whole site is indexed in /llms.txt, so agents can cite the source instead of paraphrasing it.

The rules every entry follows

Numbers carry a source link, and I fetch the source the day I write. Benchmarks are collected results, never presented as my own runs. Entries are dated and stay put; guides update in place and log every revision in a changelog. When something I wrote turns out wrong, I correct it in the open rather than quietly.

One more rule, because honesty is the whole point here: these entries are drafted with the same coding agents I write about, working from sources fetched in-session. Nothing publishes without my review. Every entry ships as a pull request that I read, edit, and merge myself, after automated checks for build health, dead links, structured data, and style. The pipeline is the practice; this site is its test bed.

No sponsors, no affiliates, no ads, no tracking. Opinions are mine, not my employer's.

Start here

The flagship isagentic coding best practices, the distillation of a year of daily agent work. If you maintain a repo that agents touch, readthe AGENTS.md guide. For a shorter taste of the format, trythe MCP servers that survived daily use.

Credentials and contact

Ph.D. in Applied Informatics, Norwegian University of Life Sciences, 2020. Publications on Google Scholar, code at github.com/hasanari, the formal CV at hasanarief.dev. Corrections are welcome and get credited:email me with the entry and the source that contradicts it.

This page updates in place; last revision 2026-07-05.