Normal's hardware runs the most memory-bound parts of large-model inference inside memory itself, rather than shuttling weights to a processor. The software that makes this programmable doesn't exist yet in any standard form. As Engineering Manager, Compiler & Runtime, you will build and lead the team that writes the core of it: the compiler that takes PyTorch models to our hardware, the runtime that executes them, and the kernels in between. This is a people-leadership role on a technical team. You will hire the team, plan its work against the hardware program, and partner with the senior compiler architect, who sets the team's technical direction. The chip is pre-silicon, so the team builds against simulators and FPGA prototypes first.
Responsibilities
Team Building: Grow the compiler and runtime team from its first engineers, set the hiring bar, and develop each engineer's growth and career; Delivery: Turn the Systems Software roadmap into plans, milestones, and staffing for the compiler, runtime, and kernels, and deliver against them; Technical Partnership: Work with the compiler architect so that technical decisions get made, staffed, and finished, and review designs and weigh trade-offs; Cross-Team Coordination: Coordinate with the hardware team (drivers and firmware) and with the serving and simulation work that depends on the compiler and runtime; Simulation-to-Silicon Continuity: Keep the compiler and runtime working against simulators, FPGA prototypes, and silicon as the hardware matures; Engineering Practices: Set up the testing, release, and review practices that make the stack reliable enough to ship to customers
Requirements
Experience managing engineers who build compilers, runtimes, or kernels for accelerators or other non-standard hardware; Track record of hiring and mentoring strong systems engineers; Enough depth in compilers and runtimes to review designs and weigh technical trade-offs with senior engineers; Experience delivering software against a hardware schedule, from simulation through bring-up; Familiarity with state of the art ML compiler stacks and with LLM inference workloads
Nice to have
Experience taking an accelerator software stack from zero to production at a startup or new hardware initiative; Experience with in-memory compute, processing-in-memory, or analog hardware interfaces; Experience shipping an SDK or software product to external customers