Principal Engineer: Annapurna Labs Training and Inference , Annapurna Labs, AWS Neuron SDK
Core
Drive AI model performance across the lifecycle (pre-training, post-training, inference) on AWS Trainium custom silicon, optimizing from kernel level to large-scale serving.
Role type
Principal Engineer (AI/ML Infrastructure & Performance)
Builds
High-performance training and inference systems for large-scale AI models on AWS Trainium/Inferentia accelerators.
Domain
Cloud Computing, AI/ML Infrastructure, Custom Silicon (ASIC)
Deliverable
production ML models
Required skills
Distributed training at scale, RL systems, C/C++, Python, AI/ML model execution, inference serving, compiler optimization, technical leadership, kernel optimization, collective algorithms, supervised fine-tuning, RLHF/RLAIF, preference optimization, inference optimization strategy, hardware-software co-design.
Preferred skills
ML model architecture (transformers, MoE, diffusion), post-training methods (SFT, DPO, reward modeling), LLM serving on custom accelerators, ML framework development (JAX, PyTorch/XLA, TRL, veRL), algorithmic optimization, numerical debugging.
Technologies
Trainium, Inferentia, Neuron SDK, PyTorch, JAX, XLA, TRL, veRL, NeMo-Aligner, TorchTitan, TorchAO, Ray, PPO, GRPO, DPO, IPO, KTO, RLHF, RLAIF.
Responsibilities
Optimize kernel performance for forward and backward paths in ML training; implement post-training alignment methods (RLHF, preference optimization); develop inference optimization strategies for cost and latency; mentor senior engineers and influence technical direction across cross-functional teams.
Seniority
Principal, hands-on IC with technical leadership