Rafael Ortega
Rafael Ortega
Artificial Intelligence
Treat models as junior colleagues: useful, fast, and in need of supervision.
Open the Special EditionAbout the Expert
Rafael designs AI workflows for product teams that cannot afford theatre. He has shipped evaluation harnesses, retrieval systems and internal copilots for companies that needed the work to survive a Monday standup. 31 Experts selected him for a rare combination: technical literacy and contempt for magic language.
- Led applied AI at two product companies, from prototype to production evals.
- Teaches operators how to specify tasks models can actually complete.
- Writes evaluation checklists before prompt libraries.
Special Edition
The Top 10
If you could give just one piece of advice to someone starting out in this discipline, what would it be?
Start with a task a careful intern could finish in twenty minutes.
What was the biggest challenge you faced in your career, and how did you overcome it?
A demo that could not survive messy real data. I learned to collect the mess first.
What is the most important “Never do this…” rule in your discipline?
Never ship a prompt you cannot evaluate.
Who inspired or mentored you, and what did you learn from them?
A statistician who asked “compared to what?” until it hurt.
What is the biggest discovery you wish you had known on day one?
The bottleneck is specification, not access to models.
What do outsiders consistently misunderstand about your discipline?
That AI “thinks.” It predicts. Your job is the thinking.
What decision looked small at the time but changed everything?
Publishing the eval set next to the feature.
What skill outside your discipline has contributed most to your success?
Interviewing users. Models fail in the sentences people actually write.
What belief did you hold when you started that you no longer believe today?
That more context is always better. Noise is a cost.
When your career is remembered, what do you hope people say you stood for?
That he made AI feel like a craft with standards.
Expert Playbook
Specify, sample, score
Write the job as a contract. Collect 20 real examples. Score them before you tune anything. Then change one variable at a time.
Digital products
Eval-first Prompt Pack
Task contracts and scoring sheets to accompany AI for Builders.
Recommended resources
A one-page eval sheet
If you cannot score it, you cannot improve it.
Media
- Evals before demos