Research Engineer (Algorithms)

Join Normal, a cutting-edge AI hardware startup, as a Research Engineer specializing in algorithms. In this role, you will develop computational methods that optimize AI inference on Normal's unique thermodynamic hardware. You will work closely with hardware and architecture teams, design numerical methods that leverage the hardware's physical dynamics, and build evaluation frameworks to characterize algorithm behavior. The ideal candidate will have a deep understanding of large model inference, stochastic systems, and experience implementing algorithms on real hardware.

• Developing computational methods for efficient AI inference on thermodynamic hardware.
• Rethinking operations like attention, memory access, and long-context decoding for stochastic analog computation.
• Collaborating with hardware and architecture teams to influence architectural decisions and shape chip capabilities.
• Exposure to analog or mixed-signal systems, in-memory compute, or non-von-Neumann architectures
• Strong programming skills in Python and at least one systems language
• Familiarity with stochastic systems, probabilistic methods, numerical analysis, or analog computation
• Experience with inference optimization: quantization, sparsity, kernel fusion, or memory-efficient attention
• Track record of taking ideas from theory to working implementation on real hardware
• Deep understanding of large model inference: attention mechanisms, KV cache, long-context decoding, memory bandwidth constraints
• Experience implementing algorithms close to hardware, not just in high-level frameworks
• PhD in machine learning, applied mathematics, physics, electrical engineering, or a related field
• Comfort reasoning from first principles about what a novel substrate can do efficiently
• Collaborative instinct and ability to work across hardware, architecture, and software teams
• Publications or open-source work in efficient inference, stochastic algorithms, or novel computing
• Experience working on hardware that did not yet exist when you joined