The files dashboard: status, review progress and due dates in one table.
The files dashboard: status, review progress and due dates in one table.
Arnita Consultants designs and executes large-scale manufacturing factories. Six architecture and engineering departments review and approve technical documents across 12-15 active projects at a time.
As the sole product designer, I designed their first document management system from research through delivery and trained the departments at launch.
Across six departments, teams compared files and called uploaders to confirm the current drawing version. With no shared view of review ownership or progress, they relied on calls to track reviews.
Review coordination via email
A complex project folder structure
I interviewed department leads, observed drawing reviews and audited shared folders to understand how teams identified versions, handed off reviews and tracked feedback.
Stakeholder interviews
I interviewed department leads to understand review handoffs, responsibilities and how they tracked progress.
Shadow sessions
I observed how reviewers found drawings, compared files and confirmed the current version with authors.
Folder audit
I examined folder structures, file names and revision records to understand how current documents were organized across departments.
I mapped the review process using interviews, shadow sessions and a folder audit to identify friction in three core tasks: finding the right version, tracking handoffs and checking revisions.
Reviewers compared three to five files and contacted authors to confirm which version to review. The lack of a clear version reference added verification effort before review could begin.
Existing workflow · Before redesign
Drawings moved through department leads or directly between teams. Without visible ownership and review status, users relied on email and phone follow-ups to understand who needed to act next.
Existing workflow · Before redesign
Both routes reach the reviewer
Reviewers pieced together comments from markups, emails and calls to check revised drawings. Separating feedback from its revision increased the effort needed to verify whether issues were resolved.
Existing workflow · Before redesign
I deferred external integrations because they introduced engineering dependencies, accepting manual uploads for the first release. I deferred live markup because departments could review at different times using persistent, version-linked comments.
I designed a browser-based platform where teams could verify drawing details, coordinate department reviews and keep feedback connected to each revision.
The system had to preserve 4 review rules:
I surfaced workload, overdue drawings and items awaiting the reviewer’s department on project cards, helping teams prioritize reviews across 6–10 concurrent projects.
I combined document names, versions, review progress, status, due dates, uploader details and comments in a scannable list. Selecting a document opened its detailed overview.
A department-level progress indicator showed completed, current and pending reviews. The drawing preview, details, comments and version history provided context for the review.
The upload flow validates files independently, allowing drafters to resolve individual issues while the rest of the batch proceeds.
The verification screen pairs the drawing preview with AI-extracted details from its title block: drawing number, revision and date. Uploader confirmation helps prevent incorrect versions from entering review, while an explained AI recommendation supports the choice of review department.
A conflicting revision label pauses only the affected file until the uploader confirms the right version. For unreadable scans, AI suggests an existing drawing to match, giving uploaders a starting point for resolving missing details. Other documents, such as PDFs and Excel sheets, get their details extracted and a suggested related drawing for the uploader to check.
Reviewers mark an area of the drawing and attach a comment to it, so feedback stays tied to a specific location and to the revision under review. Each comment carries a blocking setting, letting the reviewer decide whether it holds up approval.
Browser annotations export as a separate DWG comment overlay, letting drafters reference feedback in AutoCAD. Comment status remains in the platform, where drafters mark items fixed.
Department leads pushed to keep commenting in AutoCAD by downloading, marking up and reuploading drawings. I showed the trade-off and demonstrated this handoff with a working prototype, which won them over.
Approval stays disabled while the lead’s department has open blocking comments. The change-request dialog combines an editable AI summary with recipient selection, helping reviewers hand off feedback without AI prescribing fixes.
Synchronized side-by-side views and a revision overlay keep earlier comments visible next to what changed. When drafters upload a revision, they mark each comment fixed or not fixed, and AI highlights the changed areas, so reviewers can check the revision against its feedback.
The lead or uploader still decides whether each comment is resolved.
I trained the departments at rollout.
4 out of 6
Departments adopted within 2 weeks of launch.
180 hours
Recovered per month, across six departments.
5–7 days
Review cycles, down from 3–4 weeks.
~80%
Review requests captured entirely in-platform.
After launch, we explored four additions to support drafting, follow-ups and document retrieval.
Trust starts with a shared understanding of the problem
Using teams’ own files and review examples helped build trust in changing established habits. Now I ground stakeholder alignment in problems people recognize before discussing solutions.
Document why decisions were made
As the sole designer, I held much of the project context. I learned to record trade-offs and constraints alongside design decisions so stakeholders and engineers could understand and revisit the reasoning.
Known constraints still need user validation
I flagged the comparison view’s zoom limitation but didn’t test it early with HVAC, which contributed to launch-week friction. Next time, I’d test a constraint first with the teams it affects most.