Posted 1 month ago

ML Software Engineer

San FranciscoRemoteFull-time

AI Summary

Builds the software backbone for autonomy foundation models, designing data pipelines and training/inference workflows and integrating model architectures and simulators.

About this role

About us

We’re building the next generation of ground transportation with advanced physical AI to simplify the toughest challenges in modern freight. Our stealth team, founded by the engineers who scaled autonomous driving, is developing an entirely new vehicle platform. We move fast, stay tightly aligned between engineering and product, and focus on creating reliable, real-world autonomous systems.

What You’ll Do

Build the software backbone for autonomy-focused foundation models: design and ship multimodal data pipelines (ingest, validate, shard, package) and reproducible training/evaluation workflows (manifests, checkpoints, failure handling).

Implement and iterate on LLM, VLM, and VLA architectures; own model code paths, input/tokenization, inference runners, and output heads for downstream consumers.

Integrate and operate simulators for closed-loop evaluation; build tooling for metrics, visualization, and experiment management.

Deliver production-grade serving and inference tooling for deterministic, low-latency operation on bench/mule and eventual vehicle deployments.

Own systems from scratch: architecture → implementation → testing → documentation → iteration; raise the bar on code quality, reliability, and observability.

What We’re Looking For

Education & Experience: MS in CS/ML/Robotics or BS + ≥2 years building ML/data/evaluation systems

Software engineering excellence: Strong Python fundamentals (data structures, testing, debugging, modular design) and a track record of shipping production-quality code/APIs and reliable automation.

Pipelines → Training → Serving: Demonstrated experience building data/ML pipelines and evaluation tooling, and integrating training and inference using PyTorch, TensorFlow, or JAX.

Datasets at scale: Dataset packaging, sharding, manifest formats, and integrity checks for large multimodal datasets.

Performance & optimization: Practical work improving training/inference throughput and latency (e.g., mixed precision, efficient batching, model parallelism).

MLOps & infrastructure: Cloud storage and training workflows, containerization, CI/CD, and experiment observability (tracking, logging, metrics).

Team fit: Strong communication, collaborative with research and engineering partners, and a bias for ownership/independence in a small, fast-moving team.

Nice to have: Prior work on perception, detection, or multimodal models.