← All work
Case StudyTier AProduction2020
Municipal Mask Compliance & Health Status Reporting
Computer-vision system deployed at Zanjan Municipality using camera feeds, authorized internal records, reporting, and human review.
AITechnicalProduct
Evidence
10 cameras · ~78% observed operational accuracy
Context
2020 · Historical enterprise deployment · Maturity: Production.
Problem
The organization needed a way to monitor mask compliance during the pandemic using existing camera infrastructure.
Users
Authorized internal municipal operators and management.
My Role
Independent enterprise CV system — product/technical implementation.
Product Decisions & Responsibilities
- Designed an operational workflow that kept human review in the loop.
- Connected model output to management reporting without automating disciplinary decisions.
Technical Approach
- Built TensorFlow/OpenCV video processing and integrated with authorized identity/attendance records.
TensorFlowOpenCVVideo pipelineInternal integrationReporting
Constraints
- Real camera conditions reduced model performance.
- Health/privacy details required strict internal handling.
Product Flow
- Camera feed
- Detection/classification
- Authorized identity context
- Human review
- Management report
Outcome
- ~3 months development.
- 10 camera feeds.
- ~89% held-out test accuracy and ~78% observed operational accuracy.
What I Learned
- Offline accuracy and operational accuracy are different product realities.
- Human review matters when model outputs affect people.
Evidence Boundary
What this does — and does not — prove
No health details are exposed; ~89% is held-out test accuracy, not real-world accuracy, and the system is not described as automated disciplinary action.
Next Iteration
Extend the product only after the next decision can be tied to stronger user, operational, or business evidence rather than adding features for their own sake.