Mission
Own the LLM fine-tuning pipeline using NVIDIA NeMo Curator. Produce in-house language models that power Victoria, internal copilots, and customer-facing assistant features.
Responsibilities
- Own the NeMo Curator data pipeline — sourcing, deduplication, quality scoring
- Build SFT / DPO / RLHF training runs on our GPU cluster
- Evaluate models with internal and public benchmarks
- Partner with the Principal AI Infra Engineer on model serving
- Establish responsible-AI guardrails and safety eval
- Maintain documentation that lets a federal auditor trace a model's lineage
- Mentor ML engineers and contribute to research direction
Required qualifications
- 5+ years ML engineering; 2+ years training or fine-tuning LLMs
- Hands-on experience with NeMo, HuggingFace Transformers, or comparable
- Strong PyTorch and distributed training background
- Comfort with eval harnesses and statistical model comparison
Preferred qualifications
- Prior work on RLHF, DPO, or constitutional methods
- Background in alignment, safety, or red-teaming
- Open-source contributions to LLM tooling
- Experience with NeMo Curator specifically