Delos Data Inc - System Software Engineer

Remote - Montreal, Quebec, Canada$120k - $180k+ Equity4w ago

Remote NA Artificial Intelligence Software Engineer Systems Engineer C++Python Linux CUDA JAX

Requirements

• Strong proficiency in C++ and Python, with a deep understanding of systems programming fundamentals (memory management, concurrency, OS internals). • Proficient in a Linux development environment. • Experience with GPU programming (CUDA) and performance optimization for parallel architectures. • Familiarity with distributed AI frameworks (PyTorch, JAX, or DeepSpeed) and/or inference engines (vLLM, SGLang, Dynamo/TRT-LLM). • Hands-on experience with large-scale cluster orchestration and telemetry tools. • Education: • Bachelor's or Master's degree in Computer Engineering, Computer Science, or a related field.

Responsibilities

• Collaborate across the stack to influence the design of our foundational technology, ensuring it meets the needs of next-generation AI models. • Identify and resolve performance bottlenecks in distributed training and inference workloads through deep-dive analysis of the software-hardware interface. • Conduct rigorous performance benchmarking and characterization on multi-node clusters.

Benefits

• Level 1: I build AI pipelines or models using high-level frameworks (like PyTorch, vLLM, or DeepSpeed) in Python. I have not written low-level C++ systems software to manage custom memory orchestration or multi-node networking. • Level 2: I write C++ components for applications, but I rely entirely on standard high-level libraries (like std::thread or standard TCP sockets). I do not directly manage low-level Linux POSIX constructs or hardware-level memory boundaries. • Level 3: I write performance-critical C++ systems software. I have active experience using POSIX threads (pthreads) for strict CPU core-pinning, managing large data structures via Shared Memory (shm_open), and optimizing cluster data ingestion using asynchronous file/network I/O. • Level 4: I design core execution or communication runtimes for multi-GPU/multi-node clusters. I can confidently orchestrate lock-free multi-threaded synchronization, handle zero-copy memory transfers between host RAM and AI accelerators, and debug complex cluster latency bottlenecks at the software-hardware interface. • Recruiting Privacy Policy