Senior Product Manager, AI · Microsoft

Ravon
Lingard

I make sure AI is tested and ready before you use it.

At Microsoft, I make sure the AI models in Copilot Studio are tested and ready for customers. Outside of work, I use AI every day and build AI tools for my community. This is what I've learned along the way.

My AI stuff LinkedIn

St. Louis raised
Seattle based

Chapter 01 · What testing AI taught me

AI evals are just tests for AI.

My job is to test AI models before they reach customers. It works a lot like a test in school: someone writes the questions, the AI answers them, the answers get graded, and every new model takes the same test so we can compare report cards.

1
Write the test
Pick questions that look like the real work the AI will do.
In AI, this is the eval dataset
2
Make the answer key
Decide what a great answer looks like before anyone takes the test.
In AI, this is the rubric
3
Grade every answer
A person, or another AI acting as the teacher, scores each answer against the key.
In AI, this is the grader
4
Compare report cards
Every model takes the same test, so you can see which one is ready to ship.
In AI, this is a benchmark

Try it: grade a pop quiz

Three AI models took the same test question.

Made-up example

Question 1

A customer asks: "Can I return shoes I already wore once?"

Answer key: Returns are accepted within 30 days, unworn only. A great answer is correct, helpful and kind.

Read the three answers, guess the grades, then press the button.

What I've learned about testing AI

A test is only as good as its questions

If the questions don't look like real work, a perfect score means nothing. I look for realistic business scenarios, not trivia.

Grade the whole project, not one answer

AI agents do multi-step work. The real question is whether they finished the job: used the right tools, and recovered when something went wrong.

Retest every time

Every new model takes the test again, like a pop quiz. A model that's smarter overall can still get worse at something it used to do well.

Watch for confident wrong answers

AI can sound sure while being wrong. That's called a hallucination, and catching it is one of the most important parts of grading.

The grade isn't just right or wrong

A report card also covers safety, speed and cost. The smartest model isn't always the right pick if it's too slow or too expensive.

A person checks the grader

When AI grades AI, someone still needs to spot-check the grades. The teacher can make mistakes too.

Chapter 02 · What using AI taught me

Treat AI like your eager new intern.

I use AI every day, at work and at home. These are the habits that made the biggest difference, and the same ones I teach in my workshop.

The mindset

It drafts. You direct. You decide.

AI works fast and wants to help, but it needs direction. Let it write the first version, tell it what to change, and never let anything go out without your final say.

Better asks

Give it a role, a task, context and a format

A vague ask gets a vague answer. All four parts together look like this:

You are a social media manager for a cozy boutique bakery. Write 5 Instagram captions for our new fall menu. Our customers are busy moms who love a treat. Under 50 words each, with 3 hashtags.

My favorite trick

Let AI ask you questions first

End any prompt with one line, and the AI will ask about what it's missing before it starts. The answers get much better.

Ask me questions before you start.

Stay the boss

Fact-check, protect and edit

Double-check names, numbers and claims
Leave out passwords, customer details and financial info
Edit drafts until they sound like you
You're still the final editor

Want all of it? My workshop site has the full toolkit, prompts and a 7-day challenge: Girl, Let AI Do It For You.

Chapter 03 · How I give back

I build AI tools for my community.

Outside of work, I volunteer by building AI apps for conventions, workshops and community events. Each one taught me something I now use when I test AI. Tap a card for the full story.