Aparna Dhinakaran

Aparna Dhinakaran is co-founder and chief product officer of Arize AI, which helps teams see how models and LLM applications behave once they leave the notebook. She came from Berkeley EECS, Uber's production ML stack, and a short run at a computer-vision PhD before turning production failure modes into a company.

Pricing Models at Uber

She earned a B.S. in electrical engineering and computer science at UC Berkeley, where undergraduate research and ML coursework pulled her toward applied systems. Internships at TubeMogul (later Adobe) taught her that getting an algorithm into production often took longer than inventing it. After graduation she joined Uber's Marketplace organization, worked on forecasting and dynamic pricing, and helped build early Michelangelo-era model lifecycle and serving infrastructure. Watching models drift and break in production without software-style monitoring is the pain she later cited as the reason for observability products.

Leaving a Cornell PhD for MonitorML

She began a computer-vision Ph.D. at Cornell under Serge Belongie, then left academia after a brief stretch. With her brother Eswar she founded MonitorML, went through Y Combinator, and focused on visibility into deployed models. She later joined forces with Jason, a TubeMogul co-founder who had seen the same lab-to-production gap; MonitorML was absorbed into Arize AI, where she became chief product officer. The company grew from classical ML monitoring into LLM and agent evaluation and tracing as generative systems entered production.

Asking Whether a Model Works

Her product work centers on tracing, evaluation, and monitoring so teams catch failures before customers do. On James Le's Datacast she put the thesis bluntly: before fairness debates, teams often could not even ask "Is this model working, or how is the model performing." She argued "We won't get to the point where we continuously deliver ML as part of the software stack if we do not treat ML as an engineering discipline," and that "It is not done after you build the model. It is done when the model is in the real world." The Uber years left her comparing the missing layer to Datadog-style tools that made ordinary software operable.

Bringing Research into Customers' Hands

In the Datacast interview she contrasted research and shipping: "In research, the papers and proofs are the outputs. In the real world, you have to bring the applications you've built into the hands of customers." She speaks and writes frequently on ML ops, bias and fairness as problems that require visibility first, and the shift from notebook demos to governed production systems. She made the Forbes 30 Under 30 list and remains a regular conference voice on evaluation as generative and agent systems scale.