Managing a team of machine learning engineers is no small feat—it’s a role that demands a unique blend of technical insight, interpersonal finesse, and strategic vision. But here’s where it gets controversial: how do you cultivate a team of engineers who are not only technically proficient but also adept at collaboration and continuous learning? Vivek Gupta, an AI team manager, dives into this challenge in his talk Growing and Cultivating Strong Machine Learning Engineers (https://devsummit.infoq.com/presentation/boston2025/growing-and-cultivating-strong-machine-learning-engineers) at Dev Summit Boston (https://devsummit.infoq.com/). Gupta emphasizes that engineers thrive on feedback—not just on their coding skills, but also on their ability to work with others. And this is the part most people miss: feedback must be holistic, addressing both technical and interpersonal growth.
As a manager, Gupta stresses the importance of staying broadly informed. While senior engineers handle the deep technical dives, managers must understand the broader value of their team’s work and keep up with industry trends. Here’s a bold statement: managers don’t need to be experts in every detail, but they must be curious enough to ask the right questions and guide their team effectively. Gupta highlights three pillars for engineer growth: mentorship, data handling, and human-in-the-loop validation. Without these, even the most talented engineers can struggle to reach their full potential.
One of the most overlooked aspects of engineer development is learning time. Gupta argues that engineers need dedicated space to experiment, fail, and learn from their mistakes. But here’s where it gets controversial: how much time should organizations allocate for learning when deadlines loom? Gupta suggests regular hackathons, learning days, and lunch-and-learn sessions as practical solutions. For instance, his team participates in Microsoft’s annual hackathon and hosts internal learning sessions focused on emerging areas like AI-assisted coding.
Another critical point is encouraging engineers to ask for help. And this is the part most people miss: junior engineers often wait too long before seeking assistance, fearing they’ll appear incompetent. Gupta advocates for a culture where asking questions is celebrated, and senior engineers are trained as mentors to scale this support across the organization. Cross-team collaboration is equally vital. By sharing insights and leveraging each other’s work, teams can avoid duplication and accelerate innovation. For example, attending project design presentations from other teams can spark new ideas and foster a collaborative mindset.
Data management is another area where machine learning engineers often stumble. Gupta explains that tracking training data, test sets, and data pipelines is non-negotiable. Here’s a bold statement: inconsistent data management can derail even the most sophisticated models. Automating these processes through pipelines ensures consistency and scalability, especially during frequent retraining.
Finally, Gupta underscores the importance of human-in-the-loop validation. User feedback isn’t just about evaluating performance—it’s about refining models and ensuring they meet real-world needs. But here’s where it gets controversial: how much should we rely on human feedback in an era of increasingly autonomous AI? Gupta believes it’s essential, as it closes the loop between model output and user expectations.
In an interview with InfoQ (https://www.linkedin.com/in/gkeviv/), Gupta elaborates on these strategies. When asked about enabling learning, he highlights hackathons, sprint-end learning days, and career development sessions. For collaboration, senior engineers are encouraged to review PRs, lead design reviews, and mentor juniors. On the topic of MLOps and large language models (LLMs), Gupta notes that while LLMs introduce new challenges, the core principles of MLOps—like tracking data and building pipelines—remain critical.
Thought-provoking question for you: In a field as rapidly evolving as machine learning, how can organizations balance the need for innovation with the necessity of structured learning and collaboration? Share your thoughts in the comments—let’s spark a conversation!
