Your AI sounds sure of itself. Is it right?

I find the confident wrong answers in your chatbot, assistant or AI workflow, using your own rules, before a customer does. Five years grading AI answers for an OpenAI researcher. Now I check yours.

Worked with

NDAs welcome · Work inside your accounts · Upwork or direct invoice

Is this the decision you're facing?

"Can we trust what our AI assistant tells customers?"

Test its answers against real questions, rules and examples.

"Which AI model or setup should we choose?"

Compare the options on the same tasks and the mistakes that matter.

"Our AI vendor says it's ready. How do we know?"

Get an independent assessment with findings your team can inspect.

Services

project-based · consultancy retainer

AI Workflow Automation

Hand your repetitive writing, sorting and onboarding to AI assistants that follow your rules, and turn hours of work into minutes. A setup like this helped Dineline grow the ad spend it managed from $300K to $1M a month, with no new hires.

Fractional AI Officer

Get the AI lead your business is missing, without a full-time hire. I run your AI plan, help you choose tools and vendors, answer your team's questions, and check that every tool you pay for earns its keep.

Evaluation Design / Lead

Know your AI gets it right before customers see it. I turn your rules and documents into tests you rerun after every change, graded automatically, with a person reviewing anything the grader flags. I can also lead your team or a vendor's reviewers.

AI Stress Test

Before a customer talks your chatbot into a promise you can't keep, I get there first. I hunt for confident wrong answers, skipped steps and broken rules, and hand you each one as a test, so you catch it if it ever comes back.

Case Studies

published study

Same score. Twelve times the mistakes.

I tested 25 AI setups on 120 questions about a game's rulebook. Two scored almost the same, 99.1 and 98.9. One invented a wrong rule in 2 answers. The other did it in 24. Why a game? A rulebook works like a policy manual: clear rules, plenty of tricky cases, and answers you can check. An average score hides the mistakes that cost you. I test for those.

client work · 2023 to 2024

Dineline

A digital marketing agency serving restaurants. I built a prompt-chain tree for ad copy across Google, Facebook, Instagram and email, an onboarding assistant, and a 40-page book draft. Over that period, the ad spend Dineline managed for its customers grew from about $300K to $1M a month, with no new hires.

Erol Can Akbaba

Who you'll be talking to

In 2019 I took part in experiments at an AI research lab where, some days, my job was to lie. I played an expert trying to fool a judge, and it taught me what a convincing wrong answer needs: confidence, tidy detail, and a reader too busy to check. AI tools supply the first two all day. Busy teams supply the third. I trained as a physicist, I work from Ankara in English and Turkish, and I like finding the answer that sounds right and isn't. Bring one workflow and your rules. We'll see how it holds up.

My full background

How it works

  1. Book a 30-minute call. Tell me what your AI does and what worries you about it. If you have one answer you liked and one you didn't, bring both.
  2. Get a written scope: what I test, what you receive, and when.
  3. Get your results: a plain-language report, the test cases, and checks your team can rerun.

Prefer email? Write to erol@erolcanakbaba.com. I reply within 24 hours.

Book a complimentary 30‑min call

We can discuss your immediate problems that we can fix with AI, in your business or in your model. I will give you pointers, and then we can build on it.

Book a Complimentary Call

NDAs welcome · Work inside your accounts · Upwork or direct invoice · English and Turkish · UTC+3