Technical Documentation Generator

Last updated: Edit on GitHub

Overview

Under EU AI Act Article 11 and Annex IV, providers of high-risk AI systems must maintain technical documentation before placing a system on the market. This documentation must cover eight defined areas and must be kept up to date throughout the system’s lifecycle.

Aikraft’s documentation generator produces a structured first draft for each of the eight Annex IV sections using the information already collected during classification, plus additional inputs you provide through guided prompts. A compliance officer can go from no documentation to a complete reviewable draft in under two hours.

The generator is available on all paid plans (Starter and above). Free plan users can preview the structure but cannot generate or export content.


The Eight Annex IV Sections

Aikraft generates content for each of the following sections, as defined in Annex IV of the EU AI Act:

#SectionWhat it covers
1General descriptionSystem purpose, version, supplier details, intended use cases
2Design and developmentTraining methodology, data sources, model architecture, key design decisions
3System informationHardware and software requirements, input/output specifications
4Training and validation dataDataset characteristics, pre-processing steps, data quality measures, bias assessment
5Performance metricsAccuracy, robustness, and fairness metrics on validation and test sets
6Risk managementIdentified risks, residual risks, mitigation measures per Article 9
7Oversight and control measuresHuman oversight mechanisms, logging, audit trail configuration
8Post-market monitoring planHow the system will be monitored in production, KPIs, incident reporting process

How the AI Copilot Works

The copilot uses the structured data already in your Aikraft workspace — classification answers, system metadata, and any integrations you have connected (e.g., MLflow model cards, GitHub repository descriptions) — as its primary inputs. It generates drafts section by section rather than all at once, so you can review and refine as you go.

For each section, the copilot:

  1. Generates a draft based on available data
  2. Highlights fields where it had insufficient information and inserted a placeholder ([INPUT REQUIRED])
  3. Provides a short explanation of what regulators expect in that section
  4. Suggests follow-up questions to improve completeness

The copilot does not have access to your actual model weights, training data, or internal systems unless you connect an integration. It works from what you tell it.


Step-by-Step: Generating Documentation for a System

Step 1: Open the Document Module

From the dashboard, click your AI system’s name, then select the Document tab. If the system has not been classified yet, you will be prompted to complete classification first — Annex IV requirements vary by risk tier.

Step 2: Start a New Documentation Set

Click Generate Documentation. A panel opens showing the eight Annex IV sections. Click Generate All Sections to queue the full set, or click any individual section to generate it on its own.

Generation typically takes 30–90 seconds per section.

Step 3: Review and Edit

Each section opens in a rich-text editor with tracked changes. Fields containing [INPUT REQUIRED] are highlighted in amber. Click any highlighted field to see a contextual prompt explaining what information is needed.

You can edit any generated text directly. Changes are saved automatically and attributed to your user account in the version history.

Step 4: Check the Compliance Score

The Compliance Score indicator in the top-right corner of the Document view shows a percentage from 0 to 100. It reflects:

  • Completeness: the proportion of required fields that have been filled
  • Regulatory alignment: whether the content addresses specific obligations cited in the Act (checked via keyword and structural analysis)
  • Freshness: whether the documentation has been reviewed since the last system change or regulatory update

A score of 80 or above is considered publication-ready for internal purposes. For formal conformity assessment submissions, aim for 95+.

Step 5: Share with Auditors

Invite external reviewers (e.g., a notified body or legal counsel) with the Auditor (read-only) permission and scope them to this system. Reviewers can view your published documentation and add comments that appear in your workspace; they cannot edit content directly. The Auditor permission does not consume a paid seat.

Step 6: Export as PDF

Click Export > PDF to download a formatted PDF with your organisation’s name, system name, document version, and generation date in the header. PDF export is available on Starter plans and above.


Version History

Every time you edit a section or run re-generation, Aikraft creates a new version. To view the version history:

  1. Open the Document tab for your system.
  2. Click Version History in the left panel.
  3. Select any two versions to view a diff.

Versions are retained indefinitely and form part of your compliance audit trail.


What Triggers Re-Documentation

Aikraft automatically flags documentation as requiring review when:

  • You re-run classification and the risk tier or Annex III category changes
  • A connected integration (e.g., MLflow) detects a new model version has been registered
  • A monitoring alert indicates significant data or performance drift
  • Aikraft’s regulatory tracking system detects a relevant update to the EU AI Act, its implementing acts, or a harmonised standard

When re-documentation is triggered, a banner appears in the Document tab with a plain-language explanation of what changed and which sections are most likely affected. Existing content is not deleted — you are prompted to review and update it.