AI/ML Scientist — Quantization & Numerical Robustness

Job description

Meet Arago and the Aragonians Arago is an AI and computer hardware company whose mission is to drive the course of history forward. We do so by accelerating breakthroughs at the intersection of AI and semiconductors. Founded in 2024 by AI researchers and physicists with deep expertise in photonics, electronics, software, mathematics, and machine learning, Arago brings together a lean team of engineers and scientists from the world’s top companies and research labs. Composed of nine nationalities and operating from hubs in France, North America, and Israel, we believe in great science and fast achievements. Our work is guided by these core principles: Do great things: we deliver work we’re proud to sign our name to. High velocity: speed matters. We move quickly, one step at a time. One unit: we’re all in this together, with relationships grounded in trust, respect, and camaraderie. Arago is backed by executives from Apple, Arm, Nvidia, Microsoft, and Hugging Face, as well as prominent US and European deeptech venture firms and exited founders. What you’ll do Research how reduced precision, analog noise, and other hardware non-idealities affect modern AI models, and develop quantization and robustness techniques tailored to Arago's custom AI accelerator. The role sits at the intersection of model research, numerical analysis, and hardware/software co-design. Required Skills and Experience Strong background in mathematics, physics, computer science, or a related quantitative field, with solid foundations in numerical methods, probability, and statistics. Deep experience with ML quantization techniques, including PTQ, QAT, quantization-aware fine-tuning, mixed precision, and low-bit weight/activation formats. Experience studying the impact of numerical precision, approximation, perturbations, or hardware noise on model accuracy and stability. Strong understanding of modern model architectures, including LLMs, diffusion models, multimodal/video models, and/or world models. Ability to design rigorous, large-scale experiments and analyze accuracy/robustness trade-offs across models, layers, operators, and numerical formats. Good understanding of accelerator architecture, inference performance, memory/computation trade-offs, and the interaction between model-level techniques and hardware efficiency. Strong Python/PyTorch skills; experience with custom operators, si…