PhD Intern, Research - Model Alignment

Position Overview

Are you excited about advancing the next generation of intelligent design systems? Do you want to explore how reinforcement learning can make generative models more adaptive, diverse, and aligned with human creativity? If you're passionate about this, keep reading!

In this role, you will join Autodesk’s Research team, focusing on advancing reinforcement learning for generative design models. You will develop and evaluate new algorithms that help models self-improve, learn from feedback, and generalize across diverse generation tasks. Working closely with senior research scientists, you’ll prototype and test ideas that could lead to real-world impact in Autodesk products such as Fusion and AutoCAD, as well as publishable insights in the broader research community.

This is an exciting opportunity to contribute to Autodesk’s mission of building AI that truly understands and amplifies human design intent, helping shape the future of creative automation.

Responsibilities

Conduct research on self-play reinforcement learning (RL) methods for structured generative tasks

Design and implement RL-based alignment frameworks using PyTorch, Ray, and RayLightning for distributed training

Prototype and test algorithms that improve the validity, diversity, and stability of generated results

Collaborate with Autodesk Research scientists and engineers to integrate learned models into real-world generative workflows

Document findings and potentially contribute to a peer-reviewed publication (e.g., NeurIPS, ICLR, ICML)

Minimum Qualifications

Currently pursuing a PhD in Computer Science, Machine Learning, Artificial Intelligence, or a related field

Solid background in reinforcement learning, deep learning, language models, and generative modeling

Proficiency with Python, PyTorch, and deep RL frameworks

Understanding of distributed training or experience with Ray / RayLightning

Strong analytical and communication skills; ability to interpret experimental results and iterate quickly

Preferred Qualifications

Prior research experience in self-play RL, model alignment, or structured generative modeling (e.g., text, geometry, or graph domains)

Familiarity with CAD representations and graph neural networks

Publication or preprint experience in top-tier ML conferences (NeurIPS, ICLR, ICML, CVPR, RLC)

Enthusiasm for bridging research and real-world generative design systems

Ability to work independently while collaborating across multidisciplinary teams

About the Canada Intern Program

The 2026 Canada program runs for 16 weeks (May 4 – August 21). All internships are paid. As an intern, you will contribute to meaningful projects, be mentored by industry leaders, and participate in tech talks and other activities designed to support your personal and professional development. Our internships align with Autodesk’s Flexible Workplace approach, which is designed to meet the needs of our business while providing flexibility in support of office, remote and hybrid work preferences.

Learn More

About Autodesk

Welcome to Autodesk! Amazing things are created every day with our software – from the greenest buildings and cleanest cars to the smartest factories and biggest hit movies. We help innovators turn their ideas into reality, transforming not only how things are made, but what can be made.

We take great pride in our culture here at Autodesk – it’s at the core of everything we do. Our culture guides the way we work and treat each other, informs how we connect with customers and partners, and defines how we show up in the world.

When you’re an Autodesker, you can do meaningful work that helps build a better world designed and made for all. Ready to shape the world and your future? Join us!

Salary transparency

Salary is one part of Autodesk’s competitive compensation package. Offers are based on the candidate’s experience, educational level, and geographic location.

Diversity & Belonging

We take pride in cultivating a culture of belonging where everyone can thrive. Learn more here :