About Nightshift
Nightshift is the open-source agent platform enterprises can own. Our managed offering, Nightshift Sovereign, packages it for production: instrument the AI a customer already runs, distill it into smaller models trained on their work, and hand the result back as a balance-sheet asset on the customer's infrastructure.
About the role
The distillation pipeline that powers Sovereign is running in production for customers today. We are hiring an ML Engineer to add depth to the team that pushes it further. The work is not greenfield. It is the kind of work someone who has built systems like this before will recognize.
Representative projects
- Designing eval harnesses that catch real regressions on real customer workloads, without clean ground truth
- Fine-tuning open-weight base models on per-tenant corpora and proving the student matches or beats the teacher
- Pushing training-stack throughput and cost in the directions the team has decided matter
- Calling which workflows are worth distilling at all, and which should stay on the frontier as fallback
You may be a good fit if you
- Have hands-on ML engineering experience with production deployments
- Are deeply familiar with the open-weight model landscape and the practical knobs that move quality
- Have strong PyTorch fundamentals and work comfortably across the modern training stack
- Have built eval harnesses that catch real regressions, not toy benchmarks
- Hold cost, throughput, and quality in your head as a single moving system
Strong candidates may also have
- Shipped distilled or specialized models in production, in any domain
- Background in agent / tool-use / multi-turn evaluation
- Published research or open-source contributions in fine-tuning, distillation, or efficient inference
- Owned GPU infrastructure decisions across on-prem, cloud, or hybrid
Logistics
- Based in Miami, USA
- In-office expectation: in the Miami office at least three days a week
Candidates are encouraged to apply even without meeting all of the qualifications above.