The opportunity
SpaceX was founded under the belief that a future where humanity is out exploring the stars is fundamentally more exciting than one where we are not. Today SpaceX is actively developing the technologies to make this possible, with the ultimate goal of enabling human life on Mars.
What you'll do
Build AI systems and applications that address engineering and business: problems, working with users to understand requirements and define measurable success criteria.
Develop evaluations (evals), benchmarks, and representative datasets to: measure model and agent performance, compare approaches, and detect regressions.
Build automated graders and scoring methods, validate their agreement with: expert judgments, and investigate misleading results, data leakage, and other evaluation failures.
Create reproducible environments for training and evaluating agents,: including task definitions, tool interfaces, constraints, and feedback signals.
Integrate models and agents with engineering tools, internal services, and: company data, capturing execution traces and results to support debugging and evaluation.
Establish baselines, run experiments, and analyze failures to determine: whether changes to data, models, tools, or workflows produce meaningful improvements.
What they're looking for
- Develop data preparation, training, and evaluation workflows for problems: with limited real-world data, including targeted data collection, labeling, augmentation, and synthetic data where appropriate.
- Use Docker and Kubernetes to build and operate reliable workloads with: appropriate isolation, access controls, resource management, logging, and reproducibility.
- Collaborate with engineering, infrastructure, security, HR, and legal: partners to implement requirements for data handling, traceable outputs, auditability, and human review.
- Maintain deployed systems through testing, monitoring, ongoing evaluation,: and operational debugging, improving quality, reliability, latency, and cost based on evidence and user feedback.
