Write a model card describing intended use, performance metrics, limitations and out-of-scope uses.
Digital Transformation & Artificial Intelligence
Model Cards, Datasheets and Technical Documentation for AI Systems
Learn to write model cards, dataset datasheets and technical documentation files for AI systems that satisfy engineering, procurement and regulatory readers alike.
🗓️Duration5 training days
📚Content4 modules · 8 sessions
🎓On completionAccredited attendance certificate
About the programme
Course Overview
Most organisations discover they need proper AI documentation only when a customer's due diligence team, an auditor or a regulator asks a specific question the engineering team cannot answer from memory: what data trained this model, what its known limitations are, and what changed between one version and the next. This course teaches a practical documentation practice built on two established formats, model cards for describing a trained model's intended use, performance and limitations, and datasheets for documenting the provenance, composition and collection process of the datasets behind it, then extends both into the fuller technical documentation file that regulation such as the EU AI Act requires for higher-risk systems. Participants learn to write documentation that different readers can actually use: an engineer checking whether a model suits a new use case, a procurement reviewer assessing supplier risk, and a regulator verifying compliance, without producing three separate documents that drift out of sync. The course covers where source information should be captured during development so documentation is not reconstructed after the fact, and how to keep a documentation set current as a model is retrained. Participants leave with a completed model card, a dataset datasheet and a documentation template ready to apply to their own systems.
Expected Learning Outcomes
Produce a dataset datasheet covering provenance, collection process, composition and known biases.
Extend model card and datasheet content into a technical documentation file meeting regulatory requirements.
Capture documentation source information during development rather than reconstructing it after release.
Version documentation alongside model and dataset updates so records stay accurate over time.
Tailor the same underlying documentation for engineering, procurement and regulatory audiences.
Design a documentation template and workflow the organisation can reuse across future AI systems.
Who Should Attend
Machine learning engineers responsible for documenting trained models
Data scientists preparing datasets for internal or external use
AI governance leads standardising documentation across systems
Compliance teams preparing technical documentation for regulators
Procurement reviewers assessing supplier-provided AI documentation
Product managers explaining AI system capability and limitations to stakeholders
Course Modules
Select any module to see its sessions and points.
01Writing the Model Card
2 sessions · 8 points
Session 1Structuring Intended Use and Performance
- Describe the model's intended use cases and the contexts in which it was validated for deployment.
- Report performance metrics broken down by relevant subgroups rather than a single aggregate figure.
- State explicitly the use cases the model was not designed or tested for, to prevent misapplication.
- Reference the evaluation datasets and methodology used to produce the reported performance figures.
Session 2Documenting Limitations and Ethical Considerations
- Disclose known failure modes and the conditions under which the model is most likely to produce them.
- Document fairness testing performed and any disparities identified across demographic groups.
- Note dependencies on upstream models, libraries or services that affect the model's behaviour.
- Include contact information for reporting issues discovered after the model card is published.
02Writing the Dataset Datasheet
2 sessions · 8 points
Session 1Provenance, Collection and Composition
- Record how and from where the dataset's data was originally collected or generated.
- Describe the dataset's composition, including size, categories and any known imbalance.
- Document consent and licensing status for data included in the dataset where applicable.
- Note any preprocessing, cleaning or filtering steps applied before the dataset was used for training.
Session 2Known Biases and Recommended Use
- Disclose known biases or gaps in coverage identified during dataset construction or later review.
- Recommend appropriate uses for the dataset and flag uses the dataset is unsuited to support.
- Specify update and maintenance responsibility for the dataset going forward.
- Cross-reference the datasheet from every model card for a model trained on this dataset.
03Building the Technical Documentation File
2 sessions · 8 points
Session 1Meeting Regulatory Documentation Requirements
- Extend model card and datasheet content into the fuller documentation structure regulation requires.
- Document the system's design specifications, architecture and development methodology.
- Include risk management measures and testing evidence relevant to the system's classified risk level.
- Structure the file so an external assessor can locate required information without extensive guidance.
Session 2Capturing Information During Development
- Embed documentation checkpoints into the model development lifecycle rather than a post-hoc writing exercise.
- Assign documentation ownership to specific roles at each stage of data collection and model training.
- Use lightweight templates during development that feed directly into the final documentation set.
- Review documentation completeness at each major development milestone, not only before release.
04Versioning, Maintenance and Audience Adaptation
2 sessions · 8 points
Session 1Versioning Documentation Alongside the Model
- Version model cards, datasheets and documentation files in step with model and dataset updates.
- Maintain a changelog summarising what changed between documented versions and why.
- Archive prior documentation versions so historical claims about a model's behaviour remain traceable.
- Trigger documentation review whenever a model is retrained, fine-tuned or repurposed for new use.
Session 2Tailoring Documentation for Different Readers
- Produce an engineering-facing summary that highlights integration constraints and known limitations.
- Produce a procurement-facing summary that answers the due diligence questions buyers typically ask.
- Produce a regulator-facing extract mapped directly to the specific documentation requirements in force.
- Maintain all three views from a single underlying source so they cannot drift out of consistency.
What the participant receives
4 course modules
A structured syllabus
8 training sessions
across 5 days
32 detailed points
Applied, detailed content
Accredited attendance certificate
On completing the programme
Complete your registration
We will contact you within one business day to confirm.
Ready to start?
Reserve your seat and start building the skill.