Member of Technical Staff - ML Research Engineer, Pre-Training Data
Remote • San Francisco • FullTime
Posted 18h ago
Job Location
San Francisco
Tech Stack
Remote Work Policy
Fully remote
Employment Type
FullTime
Categories
AI Research Engineer
About the job
Liquid AI, a company spun out of MIT CSAIL, is developing general-purpose AI systems designed for efficient deployment across various targets, from data centers to on-device hardware. They focus on low latency, minimal memory usage, privacy, and reliability, partnering with enterprises in consumer electronics, automotive, life sciences, and financial services. The Pre-Training Data team is central to this mission, responsible for sourcing, filtering, synthesizing, and mixing data that will be used to train Liquid Foundation Models. As public data sources become less effective, the quality of models increasingly relies on purpose-built datasets and understanding the impact of data changes on model performance. This role treats data as a research problem, requiring close collaboration with pre-training researchers to run experiments, answer data-related questions, and build pipelines that feed into training runs.
Responsibilities
- Source web-scale and public datasets, and build pipelines for filtering, deduplication, and decontamination.
- Design data mixtures and curricula for pre-training and mid-training, and determine data inclusion for training runs.
- Build synthetic data generation systems using LLMs, structured prompting, and distillation to supplement web data.
- Conduct proxy experiments and ablations to measure the impact of data changes on model quality and identify scalable improvements.
- Develop trustworthy evaluation sets and maintain clear data lineage for reproducibility of training runs.
- Own critical pre-training data pipelines end-to-end.
- Build and improve data systems that measurably enhance model performance.
- Identify and integrate external datasets that significantly improve model capabilities.
Requirements
- Deep, hands-on experience in pre-training data work, including web-scale sourcing and curation, synthetic data generation, mixing and curriculum design, deduplication, and decontamination.
- Ability to reason about the impact of data decisions on training, including interactions with model size, architecture, and training dynamics.
- Experience designing experiments to test data changes, such as baselines, scaling ladders, small-scale proxies, and held-out loss, with clear explanations of outcomes and reasons.
- Experience building and owning production data pipelines, not just research prototypes.
- Solid ML fundamentals, including training, evaluating, and iterating on models in Python (PyTorch preferred).
- 3+ years of relevant experience with an M.S., 1+ year with a Ph.D., or 5+ years with a B.S.
Benefits
- Competitive base salary
- Equity
- Unlimited PTO
- Company-wide Refill Days
- Long weekends
- Comprehensive health coverage
- Paid leave for pregnancy, parenting, medical needs, and caring for family