Principal Research Scientist – Scaling
Leads research team advancing LLM scaling, post-training, RL, and inference efficiency. Drives innovations in optimization, distributed systems, and production integration using Python/PyTorch, with deep expertise in large-scale ML.
About the job
Responsibilities
- Lead and grow a multidisciplinary research team focused on foundational and applied AI problems, with emphasis on LLM scaling, efficiency, and systems performance.
- Define the scaling research roadmap aligned with strategic objectives, prioritizing foundation model efficiency and large-scale training/inference.
- Drive algorithmic innovations for large-scale neural network training/inference, including novel optimizers, low-precision techniques, and model adaptation methods.
- Optimize end-to-end ML systems for distributed training/RL, memory/compute efficiency via collaboration with systems/platform teams.
- Partner with product/engineering to translate research into customer-impacting capabilities.
- Foster scientific excellence, reproducible experimentation, and knowledge sharing.
- Represent research externally via publications, talks, and collaborations.
- Mentor and develop research scientists/engineers.
What You Will Do
- Define/lead research programs on foundation model efficiency (optimizer design, low-precision training/inference, scalable architectures, efficient adaptation).
- Oversee large-scale experiments, benchmarking, and trade-off evaluation (quality, latency, throughput, cost).
- Work hands-on with Python/PyTorch for research implementation, prototyping, and production integration.
- Collaborate on distributed training, parallelism, memory management, hardware utilization.
- Establish metrics/evaluation protocols for scaling research (training efficiency, inference cost, energy usage).
- Champion responsible deployment ensuring model reliability/safety.
Requirements
- Proven leadership of research teams developing novel foundation model efficiency techniques with industry impact.
- Deep expertise in generative AI, LLMs, distributed ML systems, model optimization, or responsible AI, emphasizing scaling/efficiency.
- Hands-on leadership with strong Python/PyTorch programming skills.
- Ability to translate research into scalable product capabilities.
- Excellent communication, leadership, stakeholder management skills.
Nice to Have
- Experience at systems/ML intersection (distributed training frameworks, compiler/kernel optimization, memory/compute-efficient design).
- Strong network in large-scale ML with conference service/collaborations.
- Record of research impact (top ML/systems publications, open-source contributions, deployed systems).
Skills
PyTorch, Python, LLMs, Distributed Training, Model Optimization, Low-Precision Training, Rl, Neural Networks, Foundation Models, Scalable Architectures
Similar jobs
AI Research jobs
Order.co
Boston, MASets company-wide architecture and strategy for data and applied AI, connecting governed data foundations to production intelligence and measurable business outcomes. The role requires 14+ years of experience, strong production engineering judgment, executive partnership, and hands-on delivery.
Shield AI
San Mateo, CA
Leads design, implementation, integration, and field validation of tactical autonomy software for unmanned systems and multi-agent missions. Requires extensive autonomy or robotics experience, strong C++ and Python skills, and the ability to obtain a SECRET clearance.
Ambral
New York, NYLeads the research agenda and hands-on development of replayable enterprise environments, agent evaluations, and post-training systems. The role requires deep AI research experience, a PhD or equivalent track record, and the ability to translate open-ended questions into production systems.
Shield AI
Washington, DCLeads technical direction and develops maritime autonomy capabilities for unmanned surface and underwater vehicles, including motion planning, localization, safe behaviors, and heterogeneous multi-agent collaboration. Requires deep robotics and unmanned-systems experience, strong C++/Python skills, and senior technical leadership.
Deepgram
San Francisco, CALeads Deepgram’s end-to-end TTS research program, setting technical direction, training and evaluating large-scale speech-generation models, and turning breakthroughs into production systems. The role combines hands-on technical leadership with building and developing a high-performing research organization.