Kevin Bernstein

Kevin Bernstein is Principal AI Solution Architect at UiPath, a role he took on in October 2026. In more than a decade of designing and deploying automations, his work has ranged from implementing a robotic process automation program at Intuit to helping UiPath customers decide which steps in a business process could benefit from an agent.

During his eight years at Intuit, Kevin worked on automating repeatable tasks. As a Senior Technical Account Manager at UiPath, he worked alongside strategic and technology customers to build workflows that combined automation and AI. He then became a Senior Incubation Solution Architect. In that role, he was one of several instructors for a hands-on workshop in Phoenix where participants were to build an agent prototype and connect it to a larger workflow.

Where a mortgage application waits

Kevin recommends finding the bottleneck before choosing what to automate. Process mining traces how an application moves through a workflow, including where it waits. Task mining shows what a person does within an individual step. If applications are waiting for review after someone has already extracted the financial figures, speeding up extraction will do little to shorten that queue.

In Kevin’s mortgage review example, document-processing software extracts financial figures from a borrower’s paperwork. An agent could compare those figures with the lender’s criteria and recommend how to proceed. The lender decides whether to approve the loan. Extraction and review also need separate tests. A system can read every figure correctly and still make a poor recommendation.

Giving a model one step in Flow

At AI Engineer CODE in November 2026, Kevin and Chibi Vikram are scheduled to present the UiPath team’s J Stack. The team integrated Jev as a node in Flow, UiPath’s visual workflow builder. A developer can place a model decision at a particular step and use predefined rules to determine what happens next. The team’s components handle different parts of that work. JExtract turns documents into structured data, JRank reorders retrieval results, and JEval measures how well an agent performs.