About Neuralex

A personal project about AI

Neuralex is not a company and offers no services. This site shows what runs here: the projects, how they were built and which measurements we do — only what you can recheck yourself.

What this is

Projects and technology

Neuralex is a personal project. What you read here is what has been built: working demos and measurements you can recheck.

Not everything is finished. Every project on this site has a phase — idea, in progress or live — and it says which. What (not yet) works is stated too.

Who is responsible

Ivo Donker

Ivo Donker explores what is possible with AI and, together with AI assistants, turns that into autonomously working systems. A citation graph over Dutch case law, for example: 949,000 rulings searchable in full text, 3.7 million on metadata, with a source under every answer.

Claude, an AI assistant, helps with design, code and maintenance.

A language model can write a fluent, convincing and completely wrong paragraph. That's why we only build things where every claim can be rechecked — and where we honestly state what we don't know.

Where it comes from

In legal text, an invented citation is worse than no answer: a wrong ECLI or statute article can be checked and is wrong. That is why this project's systems state the source with every answer, and why an evaluation harness reports what share of the answers actually carries that source.

The projects

What runs here

From a searchable archive of Dutch case law to an AI NOC with guardrails. Each project page states how it was built and what we measure.

Our infrastructure

Own hardware, own rules

The systems of this project run on our own hardware — no cloud provider in between, apart from the content network.

Our own homelab

A Mac Studio and a Synology NAS run the applications of this project — self-hosting in practice.

Reachable via Cloudflare tunnel

The site and applications sit behind a Cloudflare tunnel; the edge runs Caddy as reverse proxy.

GDPR from foundation to interface

Secrets kept out of code, minimal logging, no tracking. Privacy is built into the architecture, not a checkbox.

How we work

Four fixed rules

Verify before you claim

No assumptions. We read the source, check the log, and test the assumption — only then do we report.

Data is sacred

Backup before every change, nothing irreversible without confirmation. Losing data is not an option.

Actions over words

No endless decks. A working system says more than ten presentations.

Reproducible work

What we build is documented and rebuildable. No magic that only runs on one machine.

Keep looking

See the projects

Every project page describes what it is, how it was built and what we measured — with a link to the live product.