AI. Alex Ihnatsiuk

ENGINEERING SANDBOX / 01 SIMULATION MODE

Human judgment.
AI-powered quality.

Explore how I approach AI-assisted QA: turn requirements into traceable tests, inspect every agent handoff, and measure the value of human review.

Interactive local demonstration. No AI service, Azure DevOps, GitHub, or Playwright runner is connected. Your story stays in this page’s memory.
Ready. Choose a sample or enter your own story.

01 / REQUIREMENTS AGENT

Start with the story.

LOCAL RULES + TEMPLATES

20–6,000 characters. Use explicit Given/When/Then criteria. Local rules produce specific draft assertions only for resolved criteria; unsupported or ambiguous wording prompts questions.

Requirements & review questions

Run the analyzer to inspect requirements, risks, and clarification questions.

02 / TEST DESIGN AGENT

Make coverage traceable.

Rule-generated drafts require human review. No expected behavior is invented for unresolved criteria. Priorities default to P1 for demonstration and require risk review. Exports include all cases, statuses, source lines, and revision metadata.

Analyze a story, then generate test cases.

03 / MULTI-AGENT ORCHESTRATION

Every handoff, inspectable.

Four simulated agents → structural quality gate → explicit human approval → local PR preview. No remote pull request is created.

Human review checkpoint

Review the requirements, unresolved questions, draft tests, and automation blockers above. Approval only acknowledges this demonstration; it does not certify production readiness.

04 / PROOF OF VALUE

Measure the whole effort.

ILLUSTRATIVE ESTIMATES

Adjust the assumptions

Estimated comparison

Evidence, kept explicit

Sample dataStories, agent outputs, and execution counts are demonstration material. No tests have been executed here.

AssumptionsHuman labor, AI processing, review, rework, and provider cost are separate editable assumptions. Net value subtracts provider cost; setup and maintenance are excluded. Processing time is not wall-clock savings.

Verified resultsNo independently verified measurements are loaded. Requirement-link completeness does not prove extraction completeness, test quality, or human review. Workflow acknowledgment does not mark individual cases reviewed.

BUILT FOR THE NEXT CONNECTION

One interface. Real services, later.

The frontend uses an asynchronous provider interface, ready for a secure same-origin Cloudflare Worker adapter. Future AI services, Azure DevOps, GitHub, and Playwright execution belong behind that backend. Credentials never belong in browser code.

Explore my engineering work ↗