LLM Research Scientist (Pre-training & Computer Vision & Adversarial Robustness)

What You'll Do5

• Train image classifiers and generative image models from the ground up, and fine-tune open-weight language models for specific behaviors.
• Get maximum performance from limited datasets, compute budgets, and small model sizes.
• Harden models against both adversarial inputs and adversarial conversational tactics.
• Compress models via techniques like pruning, quantization, and distillation to meet strict size and latency limits while preserving accuracy.
• Troubleshoot and resolve training instabilities or convergence issues as they arise.

Requirements7

• 3+ years of hands-on machine learning research experience (including PhD research) with deep expertise in at least one of: adversarial robustness, efficient computer vision, generative image modeling, LLM post-training, or multilingual pre-training.
• Advanced proficiency with PyTorch, JAX, TensorFlow, or another major deep learning framework, and a track record of training models end-to-end.
• Solid understanding of adversarial robustness methods such as PGD-based adversarial training, TRADES, and evaluation with AutoAttack under L∞ threat models, including avoiding gradient masking.
• Experience with model compression techniques (quantization, pruning, knowledge distillation) and deploying models under hard size or latency constraints on edge or embedded devices.
• For LLM work: hands-on experience with SFT, preference optimization (DPO, RLHF, RLAIF), synthetic dataset construction, and steering multi-turn conversational behavior.
• For multilingual work: experience training tokenizers and models for low-resource or typologically diverse languages, and balancing skewed per-language data.
• A degree from a top-100 university, experience at a FAANG or similar AI organization, or a strong publication/open-source record.

Who Should Apply

You're a researcher who likes digging into messy, empirical ML problems and iterating until the model behaves. You're comfortable working independently on loosely specified research questions, and you care about making models not just accurate but robust, efficient, and deployable. You have strong coding skills in modern ML frameworks and can point to concrete projects — papers, open-source work, or deployed systems — that show your ability to train and improve deep learning models end-to-end.

Salary Insight