Every field
of a technical specification drafted by an assistant, then reviewed and signed off by the engineer who owns it.
Assistants that create, read, update and populate technical specifications for distribution engineers, from the utility's standards and the project's own data.
- Industry
- Utilities
- Segment
- Distribution engineering
- Stage
- Build Agents
- Status
- Delivered
AI-Assisted Spec Generation
The findings in three sentences.
- 01
Most of a specification is not judgement. It is lookup. That is the part worth automating first.
- 02
Flagging unknowns instead of filling them was the single design choice that won over reviewers.
- 03
Keeping the engineer as the approver kept the process inside existing quality procedures.
Client problem
What wasn't working.
- Technical specifications for distribution projects were assembled by hand from templates, standards and previous projects.
- The same values were retyped across documents, so one change meant hunting through every copy.
- Senior engineers spent review time catching transcription errors instead of engineering ones.
Dviz solution
What we built.
- Built CRUD-based AI assistants that populate each section of a specification from standards, project data and approved prior specs.
- Each populated value carries a link to where it came from, and anything the assistant could not source is flagged, not guessed.
- Engineers review, edit and approve inside the same workflow. Nothing is issued without their sign-off.
- Updates to a standard can be pushed to every open specification that references it.
A working sketch of the system, with placeholder data.
Delivered on Sorgi
How the platform carried it.
- Agent reads standards and project data without moving it.
- Approval step built into the workflow before any document is issued.
- Version history and audit log for every generated value.
Outcomes
Every field
Populated or flagged before review
1 approver
The engineer who owns the spec signs off
Next use case
Automated Power Flow Analysis
Have a problem like this one?
A backlog nobody can clear, a pilot that never left the lab, a process that only three people understand. Tell us what it is, and we'll tell you honestly whether AI is the answer.