Plate Recognizer as a Service
Problem
A civic reporting workflow involved manual license plate review. The stated pain was simple: if the plate-reading step could be automated, the team would have more capacity.
Build
A small plate-recognition service was built around OpenALPR and packaged as an OpenWhisk Docker function. The service took images and returned detected plate data for review.
What happened
The system was built quickly and handed off. The reaction was not "great, now the process is fixed." It was closer to "we are not ready for this yet." The automation exposed questions that had not been settled: who trusts the output, who reviews exceptions, what counts as acceptable accuracy, and who owns the changed workflow.
Lesson
The requested tool is not always the real bottleneck.
Why it matters now
A useful agent is not just a technical build. It needs a workflow owner, review path, trust boundary, and adoption reason. Otherwise, the system can be correct and still fail to matter.
Code
The code still lives at rawkintrevo/plates-as-a-service.