Lead AI Systems Architect - MoE Runtime & Memory Hierarchy

Lexarenterprise; San Jose (CA); Hybrid; USD 210,000 - 350,000; Full time

Job summary

Lexar Enterprise seeks a senior systems architect to own the end-to-end technical architecture for sparse MoE inference and memory management. You will convert strategic direction into concrete models, APIs, data paths, cache policies, and performance models across runtimes and hardware.

You will bridge model architecture, runtime software, OS memory management, and accelerator execution, guiding design from prototype to production while ensuring latency, reliability, and security across mobile,

Qualifications

• 15+ years in systems software, storage software, HPC, OS, ML infra, or related engineering.
• BS/MS degree in Computer Science or equivalent experience
• At least 5 years in AI/ML infrastructure, transformer inference, model serving, or accelerator software.
• Demonstrated experience architecting complex systems across multiple software or hardware layers.
• Strong C++ and Python skills with debugging concurrency, memory, I/O, and performance issues.
• Deep Linux systems knowledge (VMs, mmap, page cache, NUMA, DMA, AIO).
• Experience with NVMe/UFS/PCIe or other storage interfaces and firmware.
• Ability to translate architecture into testable requirements and milestones.

Responsibilities

• Architect the model-aware residency system with explicit lifecycle, identity, placement, quality, deadline, and security attributes.
• Design HLC policies for hot, warm, cold, and prefetched states including eviction, routing telemetry, and compression decisions.
• Develop quantitative workload models for tokens, cache hits, throughput, and bandwidth.
• Own deployment tracks for mobile/edge and enterprise/near-data with diverse storage and execution paths.
• Integrate with runtimes like llama.cpp, vLLM, SGLang, ExecuTorch, and vendor stacks; determine OS vs. hardware changes.
• Architect agent-state storage across accelerator memory, DRAM, CXL, local SSD, and remote storage.
• Design runtime-to-system APIs for expert requests, prefetch deadlines, and telemetry.
• Lead performance and correctness reviews to balance latency, energy, and cost without compromising correctness.
• Guide implementation, reviewing C++, Python, kernel, firmware, and simulator designs; mentor senior engineers.

Skills

C++

Python

Linux systems

Transformer inference

MoE routing

Performance tuning

Memory management

GPU/CPU/NPU

Software architecture

Kernel/firmware

Education

BS/MS in Computer Science

Tools

llama.cpp

vLLM

SGLang

ExecuTorch

Job description

Lexar Enterprise seeks a senior systems architect to own the end-to-end technical architecture for sparse MoE inference and memory management. You will convert strategic direction into concrete models, APIs, data paths, cache policies, and performance models across runtimes and hardware.

You will bridge model architecture, runtime software, OS memory management, and accelerator execution, guiding design from prototype to production while ensuring latency, reliability, and security across mobile,