Korstead LinkedIn
← All builds
AI visionDroneSmart glassesAzure

AI Vision for asset inspection

Checking poles, transformers and street electrical boxes usually means a truck roll for every look. Here a drone flies past first and AI says what it sees. An inspector with smart glasses follows up only where it's needed. A supervisor approves the work, and a technician fixes it. Every step becomes part of the asset's history.

Ask for a walkthrough ↗ The demo is private. Message me on LinkedIn for a live walkthrough.
The supervisor's review inbox: work orders by urgency, with the open one beside the list

How it works

Four people, one flow. Each step hands its result to the next, and nothing is typed twice.

  1. The drone flags itA detector watches the live video and flags assets. A vision model describes any visible issue.
  2. The pilot decidesThe pilot orders an inspection when the picture isn't enough, or a work order when the issue is clear.
  3. The inspector checksSmart glasses take photos and voice notes on site. The AI turns them into a work order, or the asset is marked OK.
  4. The supervisor approvesWork is reviewed in an inbox, ranked by urgency, and assigned to a technician.
  5. The technician fixes itThe repair is marked done, and the asset's record keeps the whole story.

Inside the dashboard

One web app with a view for each role, built for a laptop or a phone, in light or dark mode.

The end-to-end flow by role
The flow. Every step by role. Press one to open its screen.
The technical view: the back end of a step as a chain of boxes
Under the hood. Each step's back end as a chain you can explore, with live example data.
Insights: work in flow, on-time rings and the crew's day
Insights. Work in flow, on-time targets and the crew's day at a glance.

Built with

DJI Mini 5 ProLive video from the controller to the cloud
Ray-Ban Meta glassesHands-free photos and voice, through an Android app
Object detection + GPT-4o visionFind the assets, then describe what's wrong
Microsoft AzureFunctions, storage and containers that run only when needed