Storage Systems Architect

Location – Bangalore / Pune / Kolkata / Indore

Summary

We are looking for a hands-on Storage Systems Architect to design and build storage software for AI workloads: NVMe and NVMe-oF data paths, flash tiering for KV cache and vector data, and GPU-to-storage paths. You will work with SSD vendors, AI platform companies and customer engineering teams, lead development, and contribute to presales.

Key Responsibilities

• Architect storage data paths for AI training and inference: NVMe, NVMe-oF (RDMA/TCP) and GPUDirect Storage.
• Design and build flash tiering for inference: KV cache offload and vector indexes (DiskANN-class) on SSD.
• Benchmark and characterize storage: latency, tail latency, IOPS and bandwidth using fio, SPDK perf and AI storage workloads.
• Lead C/C++ development in the Linux storage stack or in user space (SPDK).
• Evaluate new storage technology (Gen5/Gen6 SSDs, FDP/ZNS, CXL memory) and turn it into PoCs and customer offerings.
• Contribute to presales: solution design, technical proposals and customer architecture discussions.
• Mentor engineers and lead design reviews.

Qualifications

Education

• Bachelor’s or master’s degree in computer engineering or computer science, Electronics, Electrical, (or equivalent experience).

Experience

• 10 to 16 years in storage product engineering.
• Hands-on design or coding.

Must-Have Technical Skills

• Storage product development (block, file or object), not only storage administration.
• Linux I/O stack (blk-mq, NVMe driver, io_uring) or SPDK.
• NVMe and NVMe-oF.
• C/C++ systems programming and performance analysis (perf, eBPF).
• Data protection fundamentals: replication, snapshots, RAID/erasure coding.

Good-to-Have Technical Skills

• GPUDirect Storage, RDMA (RoCE/InfiniBand).
• KV cache offload frameworks (LMCache, NVIDIA Dynamo KVBM) and vLLM.
• Vector databases and ANN indexes (Milvus, DiskANN).
• Parallel or distributed file systems (Lustre, WEKA, DAOS, Ceph).
• SSD internals: FTL, FDP, ZNS.
• CXL memory tiering; Python.