Senior Software Engineer - Storage Control Plane

Lambda•San Francisco, CA

1d•$266,000 - $395,000•Hybrid

About The Position

Lambda, The Superintelligence Cloud, is a leader in AI cloud infrastructure serving tens of thousands of customers. Our customers range from AI researchers to enterprises and hyperscalers. Lambda's mission is to make compute as ubiquitous as electricity and give everyone the power of superintelligence. One person, one GPU. If you'd like to build the world's best AI cloud, join us. In the world of distributed AI training and inference, raw GPU and CPU horsepower is just a part of the story. High-performance networking and storage are the critical components that enable and unite these systems, making groundbreaking AI training and inference possible. The Lambda Infrastructure Engineering organization forges the foundation of high-performance AI clusters by welding together the latest in AI storage, networking, GPU and CPU hardware. Our expertise lies at the intersection of: High-Performance Distributed Storage Solutions and Protocols, Dynamic Networking, and Compute Clustering and Virtualization. AI training and inference relies on petabytes of data hosted on large, high-performance storage arrays. At Lambda, the Infrastructure Storage Team’s job is to ensure that the data powering AI is fast, performant, and available across a variety of access protocols (fit for purpose). We're looking for an experienced Senior Software Engineer to join our storage team. You'll join a team responsible for developing and implementing our next-generation storage software. This role requires expertise in distributed systems, and an in-depth understanding of file, block, and object storage protocols. You'll work on building scalable and resilient storage control plane that power our AI and machine learning infrastructure.

Requirements

Bachelor's or Master's degree in Computer Science or a related field.
5+ years of experience in software development for storage systems.
Proven experience with distributed systems programming and concepts such as load balancers, data-durability, consensus algorithms, fault tolerance, and data consistency.
Strong programming skills in languages such as C, C++, Go, or Python.
Experience with Linux kernel internals and system-level programming.
Experience with one or more storage protocols (e.g. S3, NFS) and file systems such as Ceph, DAOS, or similar.
Familiarity with containerization technologies like Docker and Kubernetes and running production workloads in these environments.
Familiarity with CI/CD and QA practices for distributed systems development environments.

Nice To Haves

Experience with AI/ML workloads and the unique storage challenges they present.
Knowledge of data center networking and high-speed interconnects (e.g., InfiniBand, RoCE).
Experience with performance tuning and optimization of storage systems.
Familiarity with hardware acceleration technologies, specifically GPUs and DPUs.
Production experience with VAST Data, WEKA, DDN, Pure, NetApp, or IBM Storage Scale.
Ceph at 100 PB+ in HPC or AI environments.
CXL memory pooling, computational storage, ZNS SSDs, EDSFF.
Published or presented at SNIA SDC, FAST, USENIX ATC, LSFMM+BPF, OCP, SC, or similar.

Responsibilities

Design and build a vendor-agnostic control plane that provisions, scales, heals, and meters storage across the platforms our customers actually demand, VAST Data, WEKA, DDN, Pure, NetApp, Ceph, MinIO, and the ones that don't exist yet.
Define the internal abstraction layer that hides vendor-specific APIs, failure semantics, QoS knobs, and telemetry formats behind one declarative interface, so a new vendor integration is a driver, not a re-architecture.
Build reconciliation-loop and CRD-based orchestration (Kubernetes controllers, operators, custom schedulers) that manages capacity, tenancy, encryption domains, and placement across data centers and availability zones.
Own multi-tenant isolation end to end: namespace and subsystem partitioning, per-tenant QoS and rate limiting, credential and key lifecycle, blast-radius containment, noisy-neighbor detection.
Design the capacity and placement engine: PCIe-topology-aware, NUMA-aware, failure-domain-aware. On our platforms a drive behind the same PCIe switch as the GPU it serves beats a faster drive on a different root port, and the control plane needs to know that.
Instrument everything: SLI/SLO definitions, fleet-wide performance regression detection, and the observability pipeline that makes a petabyte fleet debuggable at 3 a.m.