Insights from Cultivating Machine Learning Engineers as a Team Manager
In his role as an AI team manager, Vivek Gupta emphasizes the importance of staying informed to effectively guide AI experts and drive the team. In his talk, "Growing and Cultivating Strong Machine Learning Engineers" at Dev Summit Boston, Gupta highlights the need for engineers to receive feedback on both technical and interpersonal skills. He stresses the importance of learning time, asking for help, and cross-team collaboration. Mentorship, data handling, and human-in-the-loop validation are key to success for machine learning engineers.
As a manager, Gupta acknowledges the need to know a little bit of everything, including applied sciences, to understand the value of the team's work. He encourages senior engineers to dive deep, while he provides ideas to keep the team moving forward. One of the primary needs for engineers is feedback, as they are fresh out of school and accustomed to grades. Gupta explains that feedback covers various aspects, from coding to interpersonal interactions and collaboration with other teams.
To nurture engineers, Gupta suggests providing them with time to learn, experiment, and practice. He also emphasizes the importance of encouraging engineers to ask questions, even when they've been stuck for a long time. Senior engineers and managers should be approachable, and engineers should be guided to seek help from those who can unblock their issues.
Collaboration is another crucial aspect. Gupta encourages engineers to interact with other disciplines and teams to foster knowledge sharing and collaboration. Senior engineers can act as mentors for juniors, and coaching them on mentorship can make it more scalable for the organization. Additionally, engineers working on machine learning in a production environment need to understand data science practices and data management specific to machine learning.
Gupta also highlights the role of human validation in the loop, where users provide feedback on model performance and modifications. He emphasizes the importance of consistent data management for training and suggests automating retraining processes through training pipelines.
In an interview with InfoQ, Gupta shares his team's approach to learning and development. They regularly host hackathons and participate in Microsoft-wide hackathons. They also have a dedicated day for learning at the end of each sprint, where they share knowledge and bring in guest speakers. Recently, the focus has been on agents and using AI assistance for coding. This allows engineers to demonstrate their new skills and experiences.
Beyond technical learning, Gupta's team provides opportunities for career development. They bring in senior speakers, offer hands-on experiences with interns, and support other teams by acting as PR reviewers or tech advisors. Collaboration among senior engineers is encouraged through knowledge sharing, design reviews, and leading learning sessions for new team members.
When it comes to MLOps, Gupta acknowledges that large language models (LLMs) present similar challenges as traditional models. Fine-tuning LLMs requires tracking the data used for fine-tuning, and they need pipelines for evaluating prompts and a library of prompts for different models. While LLMs operate differently, the learnings for MLOps still apply to ensuring a well-engineered approach for production scenarios.
About the Author
Ben Linders
