Authorium / designing an AI document editor for compliance-bound government work.

AI in regulated workflows, where "almost right" is a compliance breach.

Authorium builds workflow software for government agencies: procurement, contract authoring, grant management. The team wanted to bring generative AI into document editing without introducing the one thing regulated work cannot tolerate, plausible-sounding but unverifiable changes.

In a non-regulated context, an AI suggestion that is "mostly right" is acceptable, the reviewer catches the rest. In a government document, a single uncited paragraph can fail a compliance audit. Trust in AI is not a UX nice-to-have here, it is the prerequisite for adoption.

We partnered with Authorium to design the surface where AI assists without quietly introducing risk: clear attribution on every AI contribution, audit trails that survive a review, and notifications that separate compliance alerts from collaboration noise so the right signals are never buried.

Designed outcomes (validated in prototype testing).

• 40% faster expected document creation time for RFPs and templates.
• +40% user satisfaction in collaborative editing workflows, measured in prototype usability sessions.
• 0 compliance breaches anticipated through audit trail and access-control patterns.
• +50% trust in AI-generated content, driven by clear labeling and citation chains.

Numbers reflect Authorium's internal prototype-testing benchmarks. Production rollout metrics will replace these once live.

Every AI suggestion has a source.

In an unstructured editor, AI suggestions float free of context, the reviewer has to guess where a sentence came from. We designed the Tasks system so every change, AI-generated or human, is anchored to the specific document region it modified.

• Tasks are tied to highlighted text or a section of the document, never the document as a whole.
• Assign tasks directly within the editor, filter by priority, status, or assignee.
• Auto-notify when a task is completed or requires re-review.
• Every AI-generated edit ships with a citation chain visible in the task panel.

Compliance alerts cannot be buried in collaboration chatter.

The notification panel categorizes events by type so the signals that matter for audit, AI suggestions touching regulated content, access-control changes, document approvals, surface separately from routine collaboration updates.

• Real-time alerts for edits, approvals, and AI-generated updates.
• Categorized streams: AI suggestions, reviews, compliance alerts, mentions.
• Configurable preferences for type and frequency, per role.
• Inline contextual previews so reviewers can act without leaving the document.

An audit trail that distinguishes AI from human contributions.

Government auditors do not just need to see what changed, they need to know whether a human or a model wrote each change. We designed the Activity Log to make that distinction visible at the row level and exportable for compliance reporting.

• Chronological record of every change, comment, and approval.
• Visual distinction between human edits and AI-generated contributions.
• Per-action attribution: who, when, from which source.
• Export to compliance-friendly formats for audit submission.