Machine Learning (ML) has become an indispensable part of several End-User Programmers’ (EUPs) daily work, which increasingly includes software engineering tasks. Machine Learning End-User Programmers (ML-EUPs) without the right background face a steep learning curve and an increased risk of errors and bugs in their ML models. In this work, we designed a conversational agent named “Newton” that can act as a 24/7 expert to support ML-EUPs. The design of Newton was informed by reviewing the existing literature and identifying six challenges that ML-EUPs face and five strategies to help them. We evaluated Newton’s design by conducting a Wizard of Oz within-subjects study with 12 ML-EUPs. We found that Newton proved to help support the ML-EUPs, irrespective of whether participants finished their tasks or not. Still, Newton’s design, which featured the identified strategies, effectively mitigated the challenges from the literature. Based on participants’ interactions with Newton, we proposed six design guidelines for future conversational agents in this domain.