Systems Software Engineer

About the Company

OLIX is building the next generation of AI infrastructure. The OLIX Decode Accelerator 1 (DX-1) is the first accelerator architected specifically for decode, utilizing rack-scale co-design of logic, data movement, packaging, optics, and interconnect to enable a step change in system-level performance.

Responsibilities

• Own the Runtime & Serving Stack by designing and extending distributed inference stacks like vLLM, SGLang, or TensorRT-LLM onto DX-1.
• Scale Distributed Inference through tensor/pipeline/data parallelism, collective communication, and KV-cache management.
• Engineer for Reliability at Scale by implementing fault handling, load balancing, and observability standards.
• Drive Bring-Up by evaluating system behavior via simulation, emulation, or FPGA prototyping before silicon is available.
• Set Standards Across Teams by identifying high-impact systems problems and maintaining a high technical bar.
• Shape Direction by bringing structure to ambiguous, cross-team problems and driving strategic platform improvements.

Requirements

• Deep experience in systems software with hands-on C/C++ and strong systems fundamentals.
• Proven ownership of end-to-end systems problems, specifically extending distributed inference or serving stacks in production.
• Expertise in distributed inference at scale, including parallelism strategies and memory management.
• Fluency at the framework boundary connecting accelerators to PyTorch or JAX.
• Proficiency in whole-stack debugging, including timeline tracing and reasoning from architectural constraints.
• Bachelor’s degree or higher in Computer Science, Electrical Engineering, Mathematics, or a related field.

Preferred Qualifications

• Experience with dataflow or non-GPU accelerator architectures.
• Pre/post-silicon bring-up on custom hardware (ASIC/FPGA).
• Experience with production observability at scale (hardware counters, Prometheus/Grafana).
• Depth in HPC cluster design, high-speed networking, or distributed systems.