Autodesk

Technology

PrincipalAIResearchScientistPost-TrainingAlignment

Toronto, Ontario, Canada FULL TIME
Market Sentiment
HIGH DEMAND

Neural analysis suggests this role is
optimal for Senior candidates.

The Brief

“Principal AI Research Scientist Post-Training Alignment at Autodesk. Skills: Post-training alignment, Reinforcement learning, Foundation models, AI research. Develop novel algorithms. Improve model reliability”

What You'll Achieve.

Improve model reliability; Improve model controllability; Improve model alignment; Shape model behavior; Shape model robustness; Shape model reasoning quality; Ensure workflow completion; Achieve high annotation quality; Establish model readiness criteria; Provide go/no-go recommendations

Industry & Context.

Technology
Problems you'll solve

Model analysis; Interpretability efforts; Troubleshooting

What They're Looking For.

Must Have

Deep expertise in reinforcement learning, Fluency with post-training methods, Proven experience leading research teams, Experience designing evaluation systems, Ability to communicate technical trade-offs, PhD or equivalent research experience, Experience at a frontier model lab, Publication record at leading ML/AI venues, Background in alignment research, Experience deploying AI systems, Familiarity with large-scale training infrastructure

Nice to Have

Experience with RLHF, Experience with RLAIF, Experience with DPO, Experience with PPO, Experience with agentic systems, Experience with long-horizon reasoning, Experience with tool use, Experience with model interpretability, Experience with human-in-the-loop evaluation, Experience with scientific methodology, Experience with model readiness criteria, Experience with technical risks, Experience with model limitations, Experience with model trade-offs

What You'll Do.

Develop novel algorithms

Improve model reliability

Improve model controllability

Improve model alignment

Make architectural decisions

Address pre-training challenges

Address post-training challenges

Address system level challenges

Shape model robustness

Shape model reasoning quality

Partner with infrastructure teams

Build scalable workflows

Build reproducible workflows

Contribute to publications

Contribute to patents

Design evaluation frameworks

Evaluate long-horizon reasoning

Evaluate agentic behavior

Evaluate workflow completion

Lead interpretability efforts

Drive human-in-the-loop evaluation

Establish model readiness criteria

Provide go/no-go recommendations

Communicate technical risks

Communicate model limitations

Communicate model trade-offs

How You'll Work.

Team & Collaboration

Partner with infrastructure teams; Collaborate with academic labs; Collaborate with industry labs

Communication Scope

Communicate technical risks; Communicate model limitations; Communicate model trade-offs

Process & Methodology

Go/no-go recommendations

Full Job Description

**Job Requisition ID #** 26WD98722 **Position Overview** Autodesk's domains — architecture, engineering, construction, manufacturing, media & entertainment — provide a distinctive research environment: rich structured data, long-horizon reasoning tasks, and real-world evaluation grounded in professional workflows. Uniquely, decades of investment in physics simulation engines, CAD kernels, and computational design tools give us something most labs don't have: high-fidelity, domain-grounded verifiers that can serve as reward signals for post-training. Rather than relying solely on human preference data, we can ground reinforcement learning in the laws of physics and the constraints of real engineering. These are exactly the kinds of challenges — and assets — that make post-training and alignment research here genuinely distinctive. We publish at NeurIPS, ICML, ICLR, CVPR, and SIGGRAPH. We collaborate with leading academic and industry labs. And we have a direct line from research advances to product impact at scale. This is not a role where research sits behind a wall from engineering — you will see your work matter. **Responsibilities** * Post-training for model development — from RLHF and preference optimization to agentic systems and long-horizon reasoning * Develop novel algorithms that improve model reliability, controllability, and alignment * Make principled architectural decisions about when to address challenges at the pre-training, post-training, or system level * Design and run experiments that shape model behavior, robustness, and reasoning quality * Partner with infrastructure teams to build scalable, reproducible post-training workflows * Contribute to publications, patents, and Autodesk's external research visibility * Design evaluation frameworks for long-horizon reasoning, tool use, agentic behavior, safety, and real-world workflow completion * Lead rigorous model analysis and interpretability efforts * Drive human-in-the-loop evaluation with high ann

Free ATS check

Applying for this Principal AI Research Scientist Post-Training Alignment role?

Most applicants get filtered before a human reads their resume. See if yours makes the cut.

How to Apply on Workday

  • Workday has a multi-step form — save your progress after every section.
  • "Apply With LinkedIn" can fail or lose data; manual entry is more reliable.
  • Watch for the "Submit for Review" final step — hitting "Save" alone does not submit.
  • Job requisition numbers are useful when following up with HR by email.

ANONYMOUS · UNFILTERED

What do employees actually say about Autodesk?

Real rants from real employees. Read before you apply.

Read Company Rants →