About the job
About Us
At Mirror Physics, based in New York City, we are pioneering advancements in artificial intelligence aimed at revolutionizing scientific simulations. Our intelligent systems are designed to grasp the fundamentals of physics, significantly accelerating research and development across various technological domains. We are on a mission to create the most advanced AI platform capable of predicting experimental outcomes in chemistry and materials science, seamlessly integrating with reality through high-throughput experimental verification, thereby speeding up discoveries in biotechnology, energy, manufacturing, and more. Supported by top-tier investors and scientific authorities, we are expanding our research team at a crucial juncture in our field.
Your Role
As the lead for multimodal AI at Mirror Physics, you will be at the forefront of designing and training generative models that merge language, vision, and chemical structures. Your work will be essential in addressing some of society's most critical challenges by leveraging high-quality observational data to foster large-scale concept discovery in physics, chemistry, and materials science.
Key Responsibilities
Develop and construct multimodal architectures that integrate text, visual inputs, and chemical structures.
Curate and harmonize diverse datasets sourced from scientific literature, experimental results, and high-fidelity physical simulations.
Establish training pipelines for extensive pre-training and instruction tuning on multi-GPU/TPU clusters.
Innovate evaluation methods for cross-modal reasoning (e.g., translating text to structure, visualizing reaction pathways) and physical consistency.
Collaborate with AI, applied science, and engineering teams to integrate multimodal embeddings into our product pipeline, enhancing search functionalities, reasoning agents, and generative design loops.
Engage with the AI-for-science community through active publication and participation in conferences like NeurIPS, ICML, and ICLR.
Qualifications
Ph.D. or M.S./B.S. in Computer Science, Materials Science, Chemistry, or a related field with a strong focus on machine learning.
A minimum of 3 years of research experience working with at least two modalities (language, vision, 3-D molecular structures, point clouds, graphs).
Proficiency in Python and modern machine learning frameworks (PyTorch/JAX), along with experience in distributed training tools (CUDA, NCCL, Slurm/K8s/Ray).
Demonstrated ability to publish research findings in reputable conferences and journals.