Staff Engineer Hosur

DDN is seeking a highly experienced Staff Engineer specializing in AI Data Path & Storage to lead hands-on development and integration of advanced storage systems with next-generation AI inference pipelines. This role involves coding, prototyping, and rapidly iterating on solutions in close collaboration with architects to design and deliver high-performance data movement architectures. You will leverage NVIDIA’s NIXL (Inference Transfer Library) alongside the Infinia Data Intelligence Platform to enable ultra-low-latency, high-throughput data movement across GPU, memory, and distributed storage layers, including workloads involving KV cache management and vector database retrieval. The ideal candidate brings deep expertise in distributed storage, GPU data paths, and large-scale system optimization, with a proven track record of building and shipping production-grade AI infrastructure.

Key Responsibilities

Lead the design and implementation of high-performance data movement pipelines using NVIDIA NIXL across GPU, CPU, and storage tiers.

Architect and drive integration of DDN Infinia with GPU-accelerated inference platforms for large-scale, real-time AI workloads.

Own end-to-end optimization of I/O paths between GPU memory and storage using technologies such as NVIDIA GPUDirect Storage, RDMA, and NVMe-over-Fabrics.

Define and implement multi-tier storage architectures (NVMe, SSD, object storage) optimized for inference latency, throughput, and scalability.

Lead development of advanced KV cache management strategies, including offloading, prefetching, and persistence across distributed storage layers.

Partner with AI/ML engineering teams to optimize inference performance in frameworks such as PyTorch and TensorFlow.

Establish benchmarking frameworks and lead performance tuning efforts for storage and data movement in production inference environments.

Diagnose and resolve complex system bottlenecks across storage, networking, and GPU subsystems.

Infl Apply on Kit Job: kitjob.in/job/4n4let