Marina Petzel

Marina Petzel works at Datadog, helping developers understand whether their AI applications are healthy. Previously a machine-learning engineer at Autodesk, she developed AI features for AutoCAD and led work on automating data collection and labeling. Her work also includes teaching engineers how to use AI coding assistants: building a plan, supplying relevant context, and reviewing the result before accepting it.

From data science to AutoCAD

Petzel’s interest in data science took shape in 2016, when she was studying computer science and took Andrew Ng’s Coursera course. She earned a bachelor’s degree in computer science from Moscow State University and a master’s degree in data science from the University of Rochester. Her applied data-science experience spans construction, healthcare, oil and gas, and finance.

At Autodesk, she developed machine-learning solutions for AutoCAD’s design and drafting workflows. She also led data collection and labeling using machine-learning automation, taking responsibility for the preparation work alongside model development. Her projects included computer vision; in Autodesk’s account of its AI/ML hackathon, she described wanting to broaden that experience by exploring large language models.

Teaching brought her into direct contact with the people using her team’s work. At Autodesk University in 2025, she led two classes and a workshop with co-presenter Anum Mithani. The sessions gave her feedback from AutoCAD users on AI and machine-learning features her team had spent several years developing.

Structured workflows for AI coding

Petzel also hosted internal training on AI coding assistance at Autodesk. Her guidance on AI-assisted development treats prompting as a workflow rather than a search for better wording. She recommends several concrete practices:

• Draft and critique a plan. Have the assistant propose a plan and review it, then save the resulting plan in an instructions.md file that can be referenced throughout implementation.
• Supply targeted context. Point to specific files and lines instead of pasting large blocks of code or dumping an entire codebase into a prompt.
• Review before accepting. Treat the assistant as a junior teammate: give clear instructions, guide its contributions, and inspect its work before accepting changes.

At Datadog, her focus includes the signals engineers need to assess AI application health. Her public writing on evaluation highlights a practical concern: practitioners can recognize that evaluations are essential while remaining unsure whether they are doing enough of them.

Alongside her engineering and teaching, Petzel has shared AI and machine-learning tips and mentored early-career professionals in data science, engineering, and career strategy.