Software Engineer, Data Systems

Eventual•San Francisco, CA

2d•Onsite

About The Position

Eventual is building the infrastructure for Physical AI, focusing on Scenario Mining and Data Curation for robot fleet data. Our open-source engine, Daft, is designed for multimodal AI data, handling petabytes of video, lidar, radar, and sensor data. We aim to empower Physical AI and robotics teams to instantly find, curate, and stream the data needed to train frontier models. As an agile team, every engineer has high ownership across the stack, from compute infrastructure to data storage, querying, and model training/deployment. The Data Systems team specifically builds storage and a high-throughput data engine for petabytes of multimodal data, powering real-time indexing, distributed storage/compute, and efficient dataloading to GPUs for model training and inference. We value technical autonomy, deep execution, and a passion for solving hard distributed systems problems.

Requirements

Proven track record building resilient, high-throughput distributed systems or database engines using Rust or C++.
3+ years of experience diving deep into engine internals—such as vectorized execution, query planning/optimization, distributed task scheduling, or zero-copy networking.
High agency and adaptability to thrive in an autonomous, fast-paced startup environment building cutting-edge infrastructure for frontier robotics.

Nice To Haves

Practical exposure to scaling cloud infrastructure (AWS S3) and managing heavy-compute data pipelines.
Experience with CUDA, GPU streaming, or video decoding frameworks.

Responsibilities

Build multimodal storage (Data Lake) against modern columnar data lake formats (Apache Parquet, Apache Iceberg etc) optimized for high-dimensional video, lidar and sensor logs.
Build powerful querying capabilities for multi-stage query systems, incorporating database fundamentals like partitioning, indexing, query planning, embeddings/vector search, and LLMs/VLMs for perception-based query predicates.
Improve memory stability, throughput, and zero-copy data flow through streaming computation and line-rate CUDA tensor delivery to GPUs for dataloading.