Job Requisition ID #

26WD94771

Principal Machine Learning Engineer, 3D & Generative AI Systems

Position Overview

Autodesk is transforming the AEC (Architecture, Engineering, and Construction) industry by embedding generative AI and data-driven intelligence deeply into our products. Across AutoCAD, Revit, Construction Cloud, and Forma, we are building cloud-native, AI-powered systems that operate at the scale and complexity of real-world design and construction data.

As a Principal Machine Learning Engineer on the AEC Solutions team, you will lead the design and implementation of new machine learning models for large-scale 3D data retrieval and representation learning. Your work will focus on transforming complex geometric data—meshes, point clouds, CAD/BIM representations—into high-quality embeddings and retrieval systems that power next-generation design workflows.

This role combines deep model development, production ML systems, and technical leadership. You will architect and build end-to-end ML pipelines using Airflow and AWS, collaborate closely with researchers and product teams, and set the technical direction for how Autodesk builds, trains, evaluates, and deploys 3D-aware ML systems.3D-awar

You will report to an ML Development Manager for the Generative AI team.

Location: Remote or Hybrid (Canada or United States; East Coast preferred)

Responsibilities

Technical Leadership & Strategy

• Set the technical vision for 3D data retrieval and representation learning across Autodesk’s AEC AI initiatives

• Influence short- and long-term investments in models, data infrastructure, and ML systems

• Identify architectural gaps and scalability bottlenecks, and drive cross-team alignment on long-term solutions

Model & Algorithm Development

• Design and implement new ML models for 3D data understanding and retrieval, including geometric embeddings and multimodal representations

• Apply advanced techniques such as self-supervised learning, weak supervision, and active learning to leverage large volumes of unlabeled design data

• Optimize data representations and feature extraction pipelines for downstream model performance and retrieval quality

Production ML & Pipelines

• Architect and own production-grade ML pipelines, orchestrated with Airflow, supporting:

  • large-scale data preprocessing

  • model training and fine-tuning

  • evaluation and deployment workflows

• Build scalable systems on AWS, including integration with SageMaker and distributed training or data processing frameworks

• Establish best practices for model experimentation, versioning, evaluation, and monitoring in high-throughput environments

Data Systems & Feedback Loops

• Lead the development of intelligent data processing systems that transform unstructured 3D, text, and image data into ML-ready formats

• Own the model/data feedback loop, monitoring quality, diagnosing failure modes, and guiding iterative improvements based on real-world usage

• Collaborate with data engineers and applied scientists to ensure data quality, lineage, and reproducibility

Collaboration & Mentorship

• Work closely with AI researchers, software architects, and product teams to integrate models into customer-facing workflows

• Mentor and guide ML engineers, raising the technical bar and fostering a culture of ownership, rigor, and curiosity

• Communicate complex technical ideas clearly through documentation, design reviews, and cross-functional presentations

Minimum Qualifications

• Master’s degree or higher in Computer Science, Machine Learning, Artificial Intelligence, Mathematics, Statistics, or a related field

• 10+ years of experience in machine learning or AI, with demonstrated technical leadership and hands-on model development

• Strong expertise in deep learning architectures (e.g., Transformers, CNNs, GANs) and modern ML frameworks such as PyTorch, Lightning, and Ray

• Proven experience building new models (not just applying existing ones), especially for retrieval, embeddings, or representation learning

• Deep understanding of 3D data representations and proc