Principal AI Research Scientist – Post-Training Alignment

About the role

We are seeking a Principal AI Research Scientist to lead post‑training alignment research for foundation models in Canada. The role bridges cutting‑edge academic research with product impact, driving the development of reliable, controllable, and safe AI systems.

Key responsibilities

• Lead research and development of post‑training methods such as RLHF, RLAIF, DPO, and PPO.
• Design novel algorithms to improve model reliability, reasoning, and alignment with human and system objectives.
• Define and execute experimental frameworks to evaluate robustness, safety, and long‑horizon reasoning.
• Architect evaluation systems for agentic workflows, tool use, and real‑world task completion.
• Make principled decisions on pre‑training vs. post‑training improvements and system‑level design changes.
• Lead model analysis, interpretability, and failure‑mode investigations.
• Collaborate with infrastructure teams to build scalable, reproducible pipelines for large‑scale experimentation.
• Establish model readiness criteria and provide go/no‑go recommendations for production releases.
• Contribute to scientific publications, patents, and conference presentations.

Required profile

• Deep expertise in reinforcement learning for foundation models and post‑training alignment techniques.
• PhD or equivalent industry research experience in machine learning, reinforcement learning, or AI.
• Proven track record leading or mentoring research teams in academia or advanced AI labs.
• Strong publication history in top‑tier ML/AI venues (e.g., NeurIPS, ICML, ICLR).

Required skills

• Reinforcement learning (RLHF, RLAIF, DPO, PPO)
• Foundation model research and post‑training alignment
• Model analysis and interpretability
• Design of scalable experimentation pipelines

What we offer

• Opportunity to shape next‑generation AI systems with direct product impact.
• Collaboration with world‑class researchers and engineers.
• Support for scientific publishing and conference participation.