Carlos Esteban Rodero is a solutions engineer specializing in AI evaluation, cloud infrastructure, and cybersecurity. After helping enterprises adopt Terraform and Vault at HashiCorp, he became Braintrust’s first solutions engineer and later moved into cloud security at Wiz.
Esteban studied at Southwestern University, where he played varsity tennis. He joined HashiCorp as an intern and progressed through solutions-engineering roles in Austin and San Francisco. His work included showing organizations how Terraform and infrastructure as code can reduce cloud waste through stronger visibility and deployment controls.
At Braintrust, Esteban helped establish the solutions-engineering function and guided customers implementing evaluation, observability, and feedback systems. At AI Engineer World’s Fair 2025, he co-led a hands-on evaluation workshop and appeared alongside Notion’s Sarah Sachs and fellow Braintrust engineer Doug Guthrie in a workshop on production AI. His contributions centered on practical infrastructure for testing language-model applications:
Evaluation as production infrastructure: Define the task, dataset, and scoring function; start with a handful of useful examples; version prompts; and run evaluations through application code or continuous integration.
Calibrating the evaluator: Combine deterministic checks with focused LLM-as-a-judge scoring, compare automated judgments with expert review, and repeat trials when model variability warrants it. A poor score can reveal a defective evaluator instead of a defective application.
Offline and online evaluation: Test changes before release, score production traffic afterward, and turn real user failures into stronger datasets. Compare results with established baselines instead of treating arbitrary percentage thresholds as universal standards.
Expert review for consequential decisions: Distinguish everyday user feedback from assessment by qualified specialists, particularly in healthcare, finance, and legal work, where informed human judgment helps establish reliable expectations.
More recently, Esteban has applied his infrastructure and evaluation background to AI-powered cloud security at Wiz, extending his focus on making production systems observable and dependable.
A changelog generator provides a practical path through prompt comparisons, code-defined evals, production tracing, and human feedback that improves both the application and its judges.
How do you know a change makes an AI application better?
Notion’s evaluation workflow and a Braintrust changelog workshop show how to turn product expectations, production traces, and human feedback into repeatable tests.
▶ Watch ↗