You keep the source code, the right to change it, and a licence that never expires
I am Matthijs Breebaart, and I built Remastery. The reason I built it is written up in the first article; this page is about what it can do for you.
Remastery is a system (a workbench) for improving and converting large XML collections. I have used it on a quarter of a million documents with everything in there: namespaces, metadata, mixed content. It works. If you value provenance, accountability, and document quality then the workbench might be just the thing.
The AI connection is real. I have worked with local LLMs through Ollama, and with cloud models through OpenRouter. I have used a coding agent to help build the workbench and, more importantly, to modify it. It is all very doable, and it is very likely that these skills apply to your environment too.
So if you do have a large document collection that needs to be improved, upcycled, or converted, and you're afraid of losing track, or afraid of spending too much expensive time on things that do not warrant that, then let's have a chat.
My promise is to bring the workbench to your organisation and help you use it, customise it and extend it. We will focus on learning how to find and fix issues in your documents, deal with conflicting or outdated information, pick the appropriate level of LLM assistance, and check LLM output. After I have left, you keep the tool and the skills.
I don't want to waste your time. The goal is to quickly find out if the workbench is useful in your context, and go from there.
Fixed fee per session, quoted after the first call. It sits below common tender thresholds.
1. An online session. No charge.
An hour, online. You tell me what you need to do, and by when. I show you the system working on a real collection, with models running locally and in the cloud.
If it sounds good, we can look at a few of your own documents there and then. That might take us past the hour, which is fine.
Bring your hardest questions about the code. I will put them to a coding agent live, and you see the answers as they come.
2. A pilot. If you want proof first.
We'll focus on one real defect from your own collection: authored as an issue, tested, and applied to your documents. Small and bounded. If we go on to the engagement, the fee comes off that.
3. The engagement.
Four online sessions. By the end:
Sessions are spread across a week or two on purpose. The system is built around understanding a collection gradually, and you will learn more from trying something on Tuesday and asking about it on Thursday than from eight hours in one sitting.
From this point on, you will have a lasting deliverable: the system and its source stay with you, so you can continue to iterate and improve.
4. Hands on. Optional.
If online is too limiting, I can come to you: sit with the team, and talk to the people around the project. On-site possibility depends mostly on travel time. I am based in Australia, so I won't be able to do short stints at large distances. Travel is billed at cost.
Organisations. You have a collection, you have people who will work on it, and you have a reason to prove afterwards that the work was done right.
If you are one person with a collection and the skills to work on it, you can probably build what you need yourself. What I sell is getting a team to the point where they can do this without me.
Everything. The source, a perpetual licence to use it on your project, and permission to change it. Being able to do that is the whole point; you do not depend on a vendor's roadmap.
The workbench runs on your infrastructure. By default nothing leaves your network. The source code is around 23,000 lines of Python, which means that any reasonably advanced LLM can read it for you and give you an honest, independent assessment.
Book an hour. Tell me what you need to do with your collection, and we will look at it together.
— Matthijs Breebaart, Groundwork Data Management. Twenty years of XML, information architecture and data modelling.