AI Researcher — Distillation
Featherless AI · Remote (world)
Extracted from the posting
Researches distillation of large language models into smaller, faster models, including teacher-student, self-distillation and layer-wise methods. Runs large-scale experiments and ablations, and works with engineers to productionize results. Stack is PyTorch, and publishing in top venues is part of the role.
About the Role
We’re looking for an AI Researcher focused on model distillation to help us push the frontier of efficient, high-performance models. You’ll work on turning large, expensive models into smaller, faster, and more deployable systems—while maintaining or improving quality.
This role is ideal for someone who enjoys publishing research, working close to real systems, and seeing their ideas move from papers → code → production.
What You’ll Work On
- Design and evaluate model distillation techniques (teacher–student training, self-distillation, layer-wise distillation, representation matching, etc.)
- Research tradeoffs between model size, latency, memory, and accuracy
- Develop novel distillation approaches for:Large language modelsLong-context or specialized architecturesInference-constrained environments
- Run large-scale experiments and ablations; analyze results rigorously
- Collaborate with engineers to productionize research outcomes
- Write and submit research papers to top-tier venues (NeurIPS, ICML, ICLR, COLM, etc.)
- Contribute to internal research notes, technical blogs, and open-source projects when appropriate
What We’re Looking For
Required
- Strong background in machine learning research
- Hands-on experience with model distillation or closely related topics (compression, pruning, quantization, representation learning)
- Publication experience (conference or journal papers, workshop papers, or arXiv preprints)
- Solid understanding of deep learning fundamentals (optimization, training dynamics, generalization)
- Fluency in PyTorch (or equivalent) and research-grade experimentation
- Ability to clearly communicate research ideas, results, and limitations
Nice to Have
- Experience distilling large language models
- Work on efficiency-focused research (latency, memory, throughput)
- Experience with long-context models or non-Transformer architectures
- Open-source contributions in ML or research tooling
- Prior startup or applied research experience
Why Join Us
- Real ownership over research direction at a Series A stage
- Strong support for publishing and open research
- Tight feedback loop between research and real-world deployment
- Access to meaningful compute and production-scale problems
- Small, highly technical team with deep ML and systems expertise
Example Backgrounds
- ML researchers from academia transitioning to industry
- Research engineers with published work in model efficiency
- PhD / Post-doc graduates or industry researchers who still want to publish
