$ nav
Menu +
$ cat ~/projects/ai-claims-video-review
cd .. (go back)

Human-in-the-Loop Claims Video Review

AI-assisted review product that turns model detections into faster, human-reviewed claims decisions.

Full StackReactMLData Viz

Brief

Owned end-to-end delivery of a human-in-the-loop AI claims video review product that connects model output with reviewer judgment.

Problem

Claim managers had to review long videos while detections, evidence data, video files, and workflow context lived across separate systems. They needed one experience that directed them to relevant moments while preserving human review and feedback.

Approach

I delivered across the React product surface, GraphQL APIs, evidence integrations, automated tests, performance testing, and cross-functional rollout.

Highlights

  • Built a React, Redux, and TypeScript video interface with model-detected timestamps as interactive markers.
  • Captured reviewer adjustments, approvals, and rejections as part of the feedback loop.
  • Developed Java and Spring Boot GraphQL endpoints for AI markers, evidence data, video assets, and reviewer feedback.
  • Integrated Google BigQuery, Azure Blob Storage, and external vendor systems behind a consistent client contract.
  • Established unit, integration, and API performance test coverage with JUnit and Mockito.

Outcome

Created a single review workflow that made AI predictions explainable and actionable, reducing video screening time by 60% and overall claim review time by 40%.

Stack

React, Redux, TypeScript, Java, Spring Boot, GraphQL, Google BigQuery, Azure Blob Storage, JUnit, Mockito.