About Liquid AI
Spun out of MIT CSAIL, we build general-purpose AI systems that run efficiently across deployment targets, from data center accelerators to on-device hardware, ensuring low latency, minimal memory usage, privacy, and reliability. We partner with enterprises across consumer electronics, automotive, life sciences, and financial services. We are scaling rapidly and need exceptional people to help us get there.
The Opportunity
You'll join the Pre-Training Data team, which decides what Liquid Foundation Models learn from: which data we source, filter, synthesize, and mix, and in what order. Public data sources are plateauing, so model quality increasingly depends on purpose-built datasets and on knowing which data changes actually help.
We treat data as a research problem, not an infrastructure problem. You'll work alongside our pre-training researchers on a small team, run the experiments that settle data questions, and build the pipelines that carry the answers into training runs.
What We're Looking For
We need someone who:
Thinks like a researcher, ships like an engineer: You build the pipelines and you make the model-facing calls about what goes into them. You don't hand off either half.
Leads with a hypothesis: Before changing a dataset, you can say what it should do to the model, what you'll measure, and what could fool you, such as contamination or a noisy benchmark. You hold views on pre-training pitfalls, and you test them.
Owns problems end to end: You take ambiguous work from open question to shipped result with little direction or process, do the unglamorous parts that get it there, and don't care whose job it was.
The Work
Own a part of the pre-training data stack end to end, for example a quality filter, a dedup or decontamination pass, a synthetic or distillation data source, or a mixture or curriculum experiment, from question to shipped result.
Design experiments that show whether a data change helps: baselines, small-scale proxies, scaling ladders, held-out loss and targeted evals. Report null and negative results as clearly as wins.
Keep the loop scientifically valid: check for contamination and benchmark leakage, watch for benchmark noise, and flag when a quality gain costs diversity.
Build and run production pipelines that feed real training runs at scale, and make them reproducible and traceable.
Work with pre-training, post-training and eval teammates to understand what they need from data, and propose the next most useful thing to work on.
Experience & Capabilities
What you'll need:
Hands-on depth in at least one substantial area of pre-training or mid-training data, for example web-scale curation, quality filtering, dedup or decontamination, mixtures or curricula, or synthetic data and distillation. We don't expect mastery of every technique.
Evidence you've connected a data change to a model outcome, through an experiment you designed and can explain, including what you'd do differently.
Strong Python, experience with large-scale data processing, and practical experience training or fine-tuning models (PyTorch preferred).
Agency paired with judgment. For example, you spotted a problem nobody assigned, explained why it mattered, and fixed it. Or you killed your own approach when the data said so. Or you chose not to build something and saved the team time.
A clear account of your own contribution: what you decided, what tradeoff you made, and what changed because of you.
Pragmatism: you size experiments to the compute, time and team you have, and build just enough process.
Helpful, not required:
Experience with small or efficient models and with how architecture changes should affect data strategy.
Long-context data, knowledge distillation, or eval-framework work.
Open-source data tools or datasets with real adoption, or published data research.
What Success Looks Like (Year One)
You own a critical pre-training data pipeline end to end
Data systems you built or improved have measurably moved model performance
You've identified and integrated at least one external dataset that moved the needle
What We Offer
Upstream of every model: Pre-training data shapes every Liquid Foundation Model. On a small team, the mixtures, filters, and synthetic data you build go straight into the models we ship.
Compensation & Equity: Competitive base salary plus equity, so you share in what we build. Final offers reflect skills, experience, and location.
Time Off: Unlimited PTO, plus company-wide Refill Days, long weekends when the whole company rests together.
Health & Wellbeing: Comprehensive health coverage, with details varying by location.
Paid Leave: Paid leave for pregnancy, parenting, medical needs, and caring for family.