We’re looking for a Machine Learning Engineer focused on model distillation to help us build smaller, faster, and more efficient models without sacrificing quality. You’ll work at the intersection of research and production—taking cutting-edge techniques and turning them into systems that scale.
This is a hands-on role with real ownership: you’ll design distillation pipelines, run large-scale experiments, and ship models used in production.
Design and implement knowledge distillation pipelines (teacher–student, self-distillation, multi-teacher, etc.)
Distill large foundation models into smaller, faster, and cheaper models for inference
Run and analyze large-scale training experiments to evaluate quality, latency, and cost tradeoffs
Collaborate with research to translate new distillation ideas into production-ready code
Optimize training and inference performance (memory, throughput, latency)
Contribute to internal tooling, evaluation frameworks, and experiment tracking
(Optional) Contribute back to open-source models, tooling, or research
Strong background in machine learning or deep learning
Hands-on experience with model distillation (LLMs or other neural networks)
Solid understanding of training dynamics, loss functions, and optimization
Experience with PyTorch (or JAX) and modern ML tooling
Comfort running experiments on multi-GPU or distributed setups
Ability to reason about model quality vs. performance tradeoffs
Pragmatic mindset: you care about shipping, not just papers
Experience distilling LLMs or large sequence models
Experience with inference optimization (quantization, pruning, kernels, etc.)
Familiarity with evaluation for language models
Open-source contributions or research publications
Experience in early-stage or fast-moving startups
Work on core model quality and cost efficiency—not side projects
High ownership and direct impact on product and roadmap
Small, senior team with strong research + engineering culture
Competitive compensation + meaningful equity
Remote-friendly, async-first environment
We enable serverless inference via our GPU orchestration and model load-balancing system. We unlock fine-tuning by enabling organizations to size their server fleet to throughput needs, not number of models in the catalogue.
See it in action on our public cloud, which offers inference for 10k+ open weight models.
