← All work

Systems / 2025—26

OMAD / Inspector.

An analyst-run image review path that carries meter-photo evidence from intake to a human inspection decision.

PracticeVisual inspection workflow · applied AI
My partProduct workflow, model comparison, data intake, inspection policy and review interface.
Explore02 visual assets
Inspector archive and recorded inspection results
01 / OMAD / InspectorInspector archive and recorded results · Open original ↗
01 / Context

The problem behind the work.

Inspection teams need to separate what a photo visibly supports from what requires a retake, more context or an on-site judgment.

02 / Contribution

What I built.

  1. 01

    DMS ZIP and manifest intake with image provenance, document-ID joins, policy versions and review-ready exports.

  2. 02

    Visible-sign observations, reason codes, nullable scores, retake routing and an Inspector queue for adjudication.

  3. 03

    Model comparisons against technician-reviewed cases and MLflow-linked evaluation work.

03 / Where it stands

The score expresses inspection priority, not fraud probability. Inspectors make the final call. A public interim calibration report is available; the full product requires sign-in.

Technology and methods

PythonDjangoNext.jsPostgreSQLVertex AICloud StorageMLflow
Next project / Web & designOmnisight