Roman Horodnichev

Your AI agent, tested before launch.

I find where your chatbot or AI agent gives wrong answers, leaks data or breaks its rules — before your customers do.

Get a test plan

Roman Horodnichev · QA automation engineer · Spain

Process

From first call to a tested agent.

01

Brief

You show me the agent and what it must do.

→ scope + fixed price
02

Test

I attack it by hand and with automated tests.

→ list of failures
03

Report

Every problem with proof and priority.

→ report you can act on
04

Re-test

You fix. I check again and confirm.

→ ready to launch

What I check

Four things that go wrong.

  • Answers facts only from your data
  • Actions right tool, right inputs
  • Safety prompt injection, data leaks
  • Stability same quality after updates
$ agent-checks booking-bot # example
✓ answers from the real price list
✓ refuses to show other customers' data
✓ ignores "forget your rules"
✗ books a table for 25:00
→ reported · fixed · re-tested ✓
✓ hands off to a human when unsure
5 passed · 1 fixed

Services

Fixed scope, fixed price.

CHECK

Pre-launch check

Full test of your agent, including attacks and data-leak attempts.

PACK

Regression tests

Automated tests that run after every prompt or model change.

WATCH

Monthly monitoring

Scheduled checks and a short report every month.

BUILD

AI automation

n8n workflows and agents, delivered with their tests.

Tell me what your agent does.

hello@aiamqa.com

I reply with what I would test and a fixed price.