Synthetic Data

About Fleet

Fleet studies how environments produce intelligence. We believe intelligence is an emergent property of environmental pressures — the environment determines what capabilities develop, what behaviors survive, and what "good" looks like. We work with frontier labs on post-training across modalities — building benchmarks that expose where frontier models break, training recipes that close those gaps, and scalable oversight for long-horizon agents. Backed by Sequoia Capital, Menlo Ventures, BCV, and SV Angel.

About The Role

We're looking for a Member of Technical Staff to own the synthetic data that powers our training environments. Each environment needs a purpose-built dataset — deeply understood, carefully constructed, and continually refined. This is not a data platform or data pipeline role. You'll go deep on individual datasets, treating each one as a product with its own quality bar, iteration cycle, and domain nuances.

What You'll Work On

• Build and own synthetic data products for training environments
• Design pipelines and tooling for generating rich, structured synthetic documents with deep context across multiple artifacts
• Design agentic hillclimbing mechanisms to improve data realism. Use systems-level thinking to improve consistency of data.
• Define and enforce data quality standards for each dataset — understand what "good" looks like for each domain and iterate until it's right
• Collaborate with environment and task-writing teams to ensure data products serve the exact needs of each training environment
• Develop scalable approaches combining code generation for structure with LLM research agents for domain-specific detail (medical, insurance, financial, etc.)

How We Work

• You own your data seeds end-to-end — from understanding the domain to shipping a production-quality data product ready for use in an environment
• Direct collaboration with frontier labs and internal research teams

What We're Looking For

• Experience building a dataset that was the core product — not serving other teams' data, but owning and refining a dataset as the thing customers or systems depend on (e.g., Mapbox places/geometry, knowledge graphs, domain-specific corpora)
• Comfort going deep on a single domain and iterating through many refinement cycles to get it right
• Applied research background in synthetic data generation, agentic data construction, or domain-specific data pipelines

Location

On-site in San Francisco or New York. Highly competitive salary + equity.