Founding Machine Learning Engineering Lead (Model Training) – Superintelligence Lab

About Us

We are an early-stage superintelligence research lab pursuing a fundamentally new approach to adavancing AI systems, one grounded in scientific rigor and a realistic path toward superintelligence. Our lab was founded by seasoned industry leaders and internationally recognized academic researchers at the professor level. Learn more about our mission and team at https://eqe.ai.

As part of the founding machine learning team, you will work closely with the Head of AI Algorithms to architect, build, train, and scale the training infrastructure behind a new class of models that move beyond conventional transformers.

This role is deeply technical and execution-focused, sitting at the crossroads of large-scale training, systems engineering, and novel model implementations. You will be responsible for the core engineering systems that allow looped and recurrent language models to scale efficiently, enabling step-function improvements in reasoning, efficiency, and real-world deployment.

What You Will Do

• Architect and implement large-scale training systems (~500B parameters) for non-standard LLM architectures, including looped transformer models.
• Design and build parallelization and pipelining strategies to overcome back-propagation-through-time bottlenecks in recurrent models.
• Lead performance engineering efforts across the training stack, including memory optimization, kernel efficiency, and parallel execution.
• Work deeply with JAX (preferred) and/or PyTorch, optimizing distributed training across multi-node, multi-accelerator setups.
• Implement and optimize CUDA kernels where necessary to unlock performance beyond existing frameworks.
• Scale training to multio billion-parameter-class models, leveraging tensor, pipeline, and parallelism.
• Enable efficient distillation of large models into looped architectures and support fine-tuning on reasoning-focused datasets.
• Collaborate closely with the Head of AI Algorithms to translate new algorithmic methods into scalable, production-grade training systems.

Who You Are

• A senior-level applied ML engineer with deep hands-on experience training large language models (≥1B parameters).
• Strong background in distributed training, including tensor parallelism, pipeline parallelism, and performance bottleneck analysis.
• Comfortable working close to the metal: CUDA programming, memory optimization, and kernel-level performance tuning.
• Able to reason about trade-offs across compute, memory, throughput, and model quality.

Why Join Us

• High impact: This is a core early stage role. You will help define not just how we train models, but what becomes possible.
• Work with world-class talent: You will collaborate closely with some of the world's leading researchers and academics in AI.
• Real frontier work: We are not optimizing today's transformers, we are building the training systems that enable fundamentally new classes of models.
• Right-time entry: We are immediately pre-funding, offering a rare opportunity to join at maximum leverage, before scale-up and institutionalization.

Compensation & Equity

• The goal at the point is to assemble an elite ML engineering leadership team as we fundraise. You may remain at your current job until fundraising close.
• Meaningful equity ownership and a top-of-market salary, commensurate with founding-team responsibility and long-term impact.
• We are explicitly looking for individuals who are motivated by ownership, mission, and long-term impact.

Closing Note

Success in this role will not only help shape our company's future but also impact the future of humanity at large. If you are excited by the idea of building the systems that make next-generation intelligence possible, this role offers a rare opportunity to work at the frontier of AI with exceptional autonomy and responsibility.