• Are you looking for guidance on how to create technical documentation for your AI/ML system and/or its constituent models and datasets?
• Should your AI/ML system documentation provide value to a wide variety of user personas (e.g. managers, system users, developers, researchers, regulators, etc.)?
• Does your AI/ML system need to comply with the EU AI Act, or are you developing a model or dataset that you want to be used as part of an AI/ML system that needs to comply with the EU AI Act?

If you answered yes to any of these questions, TechOps is for you.

TechOps, published in AIES 2025 (see citation details below), is a set of documentation templates and examples designed to help technical teams and companies document their AI/ML applications, models, and datasets. TechOps is also the first set of documentation templates that we know of to completely map its sections to the EU AI Act, for developers and providers that need to comply with this regulation that first came into force in August 2024.

While rendering agnostic content is the primary contribution of TechOps, content rendering is almost always needed to make such content useful to a real world audience. For documentation developers that do not already have their own systems of documentation rendering, we provide a blueprint for rendering TechOps Documentation. This blueprint is designed to make TechOps Documentation readable by a wide variety of reader personas (technical users, business analysts, regulators, etc.), by rendering these complex documents with a side navbar so readers can quickly drill down to find the information they most need to see, without losing track of the big picture (see screenshot below).

See the TechOps website to dive in deeper and/or explore our templates and examples.

If you use and parts of this work, we kindly ask that you cite our paper:

@article{lucaj2025techops,
   title     = {TechOps: Technical Documentation Templates for the AI Act},
   author    = {Lucaj, Laura and Loosley, Alex and Jonsson,  H{\aa}kan and Gasser, Urs and  van der Smagt, Patrick},
   url       = {https://ojs.aaai.org/index.php/AIES/article/view/36663},
   DOI       = {10.1609/aies.v8i2.36663},
   journal   = {Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society},
   pages     = {1647-1660},
   year      = {2025},
   month     = {Oct.}
}

Developing TechOps Documentation is easy, it's just markdown! TechOps Documentation is, on purpose, rendering agnostic, so it can be integrated with any existing documentation systems as needed. However, we also provide an mkdocs based blueprint for rendering TechOps Documentation.

Follow these instructions to get started.

• Python 3.12 or later is installed (we've tested this with Python 3.12, but it likely also works with older versions)
• Optional, but recommended, uv is installed
• You are in your own fork/copy of this documentation repo, where you can create your own documentation.

uv run mkdocs serve

Assuming you didn't change any documentation in the docs/ folder, your first rendering via mkdocs serve will render a local version of the TechOps Website. Now you are ready to replace the examples with TechOps documentation of your AI/ML system. Happy documenting.

Mkdocs websites can be deployed like any other webapp, but some online sites provide out-of-the-box online deployments. For more information on deploying, see the official mkdocs deployment documentation.