Location: Milan, Lombardy, Italy
Job Type: Full-time
Salary: $80K - $140K
Equity: 1.00% - 2.00%
Experience: Any (new grads ok)
About the job
You'd be our first researcher, on models that keep getting better after they ship.
Lucebox builds a local AI workstation and the open source engine that runs models on it. The engine is Apache 2.0, and every benchmark goes on our blog with its setup. We're backed by Y Combinator.
A few examples of your responsibilities
Research. Take open problems from idea to result, with experiments you design and run yourself.
Training. Post-train and fine-tune open models end to end on our own hardware.
Evaluation. Build the measurements that tell us whether a change made a model better.
What we look for
Research record. A PhD in machine learning, finished or close to it, or a record that matches one, with first-author papers at venues like NeurIPS, ICML, ICLR, ACL or EMNLP.
Training. You've post-trained models yourself and know where the GPU hours went.
Engineering. You're fluent in PyTorch and will read a kernel when speed is the problem.
Conviction. You want your research running on machines people own.
Good to have. Research code or models you've released that other people use.
What we offer
Pay and equity. €70,000 to €120,000 gross a year (based on experience), plus 1% to 2% equity.
A founding role. You set the research agenda with the founders and work with them directly.
Research that ships. What works goes into our open source engine and onto every Lucebox.
Commitment and location
Full time, hybrid in Milan, Italy.
The selection process
We look at what you've built and published, whatever your title. Everyone goes through the same steps:
An intro interview, 30 minutes with a founder.
A technical interview on machine learning and your research.
A behavioural interview on how you work.
A technical project on a small research question.
The final decision, within a week of the project.
Bonus
A pull request to our open source engine counts in your favour, including one you sent before. It's optional. Put the link in the form and we'll read it before the first call: github.com/Luce-Org/lucebox
If you like the role but your CV doesn't look like a perfect match, apply anyway.
Summary
You'd be the first researcher at Lucebox, taking open problems from idea to result, post-training and fine-tuning open models end to end on company hardware, and building evaluation measurements. Requires a PhD in machine learning or equivalent record with first-author papers at top venues, plus hands-on PyTorch and model training experience.
Responsibilities
Research: take open problems from idea to result with self-designed experiments; Training: post-train and fine-tune open models end to end on own hardware; Evaluation: build measurements of model improvement
Qualifications
PhD in machine learning finished or close to it, or a record that matches one, with first-author papers at venues like NeurIPS, ICML, ICLR, ACL or EMNLP; post-trained models yourself; fluent in PyTorch; want research running on machines people own
Education requirements
PhD in machine learning, finished or close to it, or a record that matches one, with first-author papers at venues like NeurIPS, ICML, ICLR, ACL or EMNLP
Experience requirements
Any (new grads ok); post-trained models yourself; fluent in PyTorch
Interview process
Intro interview 30 minutes with a founder; technical interview on machine learning and research; behavioural interview; technical project on a small research question; final decision within a week of the project