Dataloading Systems Engineer - Multimodal AI (4d/w)

Eventual; San Francisco (CA); On-site; USD 150,000 - 250,000; Full time

Benefits offered by this job

In‑person, tight-knit team — 4 days/wk

Competitive compensation and startup‑s

Catered lunches and dinners

Commuter benefit

Team-building events

Health, vision, and dental

401(k) plan with match

Job summary

Eventual in San Francisco is seeking a Systems Engineer for the Dataloading team to turn multi-petabyte video corpora into tensors on the GPU, driving the data path from object storage through NVMe, cache, and into device memory.

You will optimize for peak bandwidth on NVMe and memory hierarchies, align with Vera Rubin roadmap, and collaborate with labs and partners to push MFU end-to-end using CUDA, Rust/C++, and SLURM.

Qualifications

• Strong systems-level performance focus and flamegraph mindset.
• Proficiency with Rust, C++, or C and modern OS concepts.
• Experience with IO and storage subsystems (NVMe, memory, network).
• Familiarity with GPU data pipelines and HPC tooling (SLURM, CUDA).
• Contributions to open-source systems in Rust or C++.
• Experience with GV/FP architectures and high-throughput data paths.

Responsibilities

• Design and build the video-native dataloader, returning tensors to the GPU.
• Profile and optimize the data path from object store to device RAM.
• Saturate latest hardware (B200, GB200, NVL72) on real training jobs.
• Own performance benchmarks against baselines and prior numbers.
• Collaborate with labs to land the loader in training stacks.
• Work across Storage Infrastructure and model-output ingestion paths.

Skills

Rust

C++

C

Operating systems

Memory hierarchies

Tools

SLURM

Kubernetes

CUDA

PyTorch DataLoader

Job description

Eventual in San Francisco is seeking a Systems Engineer for the Dataloading team to turn multi-petabyte video corpora into tensors on the GPU, driving the data path from object storage through NVMe, cache, and into device memory.

You will optimize for peak bandwidth on NVMe and memory hierarchies, align with Vera Rubin roadmap, and collaborate with labs and partners to push MFU end-to-end using CUDA, Rust/C++, and SLURM.