Back to Portfolio

AI Workflow

AI-assisted Code Review Workflow

A workflow for using AI to inspect requirements, identify risks, and create review checklists for pull requests.

Context

AI can speed up software development, but without a process it can also create code that is hard to trust. This project models a workflow where AI acts as a secondary reviewer, not a replacement for engineering judgment.

Workflow

I split the review into four steps:

  1. Summarize the requirement and the change scope.
  2. Ask AI to list risks across security, data, UX, and maintainability.
  3. Compare the AI checklist with existing tests.
  4. Let the engineer confirm, adjust, and document the final decision.

Technical points

Prompts follow a fixed format: context, diff summary, expected behavior, constraints, and checklist output. For .NET backends, the checklist focuses on validation, authorization, transactions, async flow, and logging.

Result

The workflow reduces missed risks in reviews, especially for small changes that touch business logic. Most importantly, AI creates better questions while the engineer still makes the final decision.