ML Engineer (ONSITE IN SF)

<div style="box-sizing: border-box; color: rgb(51, 51, 51); text-align: center;"><strong style="box-sizing: border-box;"><span style="box-sizing: border-box; color: rgb(184, 49, 47);">Dear applicants, please keep in mind that applications without provided salary expectations and active LN profile will not be considered.&nbsp;</span></strong></div><div style="box-sizing: border-box; color: rgb(51, 51, 51); text-align: center;"><strong style="box-sizing: border-box;"><span style="box-sizing: border-box; color: rgb(184, 49, 47);">Hope for your understanding.</span></strong></div><div style="box-sizing: border-box; color: rgb(51, 51, 51); font-family: Oswald, sans-serif, ;"><br></div><div><strong>Location: San Francisco, CA (In-person)<br>Employment Type: Full-Time<br>Equity: 0.5% – 1%<br>Visa: Not available<br>Experience: 1+ years (exceptional new grads welcome)</strong></div><div><br></div><div style="text-align: justify;">We are hiring ML Engineers to implement research ideas reliably and operate full training pipelines end-to-end. This is not a research-only role. This is research-engineering at scale. A seed-stage research-driven ML company focused on mechanistic understanding of model architectures and optimizers.</div><div><br></div><div><strong>The team studies:</strong></div><ul><li>Optimizer–architecture co-design</li><li>Orthogonalized optimizers and manifold-based training</li><li>Sparse attention mechanics</li><li>Data-efficient reasoning models</li><li>Learning dynamics in data-sparse regimes</li></ul><div>The environment blends academic rigor with industrial compute and speed. The team is deliberately long-term oriented and avoids premature commercialization pressure.</div><div><br></div><div><strong>You will:</strong></div><ul><li>Translate research papers into working PyTorch/JAX implementations</li><li>Run distributed transformer training</li><li>Debug divergence and instability</li><li>Optimize throughput</li><li>Build full pipelines (data → training → evaluation)</li><li>Reason about learning dynamics and architecture tradeoffs</li><li>The bar is slope and research intuition, not years.</li></ul><div><br></div><div><strong>What You’ll Own</strong></div><ul><li>Reliable implementation of novel architectures</li><li>Distributed transformer training at scale</li><li>Training stability and performance debugging</li><li>Evaluation frameworks</li><li>Optimization reasoning alongside researchers</li></ul><div><br></div><div><strong>Must-Have Requirements</strong></div><ul><li>Strong PyTorch or JAX proficiency</li><li>Hands-on transformer training experience</li><li>Experience with distributed training setups</li><li>Debugging divergence and instability</li><li>Ability to read and implement research papers</li><li>Research intuition around optimization and learning dynamics</li><li>High growth slope</li></ul><div><br></div><div><strong>Nice to Have</strong></div><ul><li>Megatron-LM, DeepSpeed, xformers</li><li>End-to-end pipeline ownership</li><li>Research-engineering team experience</li><li>Mathematical depth (optimization, information theory, etc.)</li><li>Competitive programming / theory-heavy background</li></ul>