Senior Research Engineer, LLM Training & Post-Training

What We're Looking For

We're are looking for an experienced Senior Research Engineer who has built, trained, and optimized modern transformer-based language models to join our Research Engineering function at Lightning.

This role will focus on advancing how large language models are trained, fine-tuned, evaluated, and deployed across Lightning AI's platform and real-world customer workloads. It will work across model training, post-training, PyTorch, distributed systems, and AI systems engineering to improve model quality, training efficiency, and developer productivity while collaborating closely with researchers, infrastructure engineers, and customers.

We're looking for someone who enjoys turning cutting-edge research into production systems. You have deep experience training and improving transformer-based language models, strong software engineering fundamentals, and a passion for solving difficult problems across model training, evaluation, and AI systems. Rather than building applications on top of existing models, you're motivated by improving the models themselves and the systems that power them. Our work spans models that power the Lightning AI platform, customer-specific model workloads, and research that translates into reusable training and platform capabilities.

This role is hybrid with a minimum of 2 in-office days per week in San Francisco, Seattle, NYC, or London, with fully remote work considered for candidates outside of our office hub locations. All employees participate in occasional team and company offsites.

What You'll Do

• Design, build, and optimize training and post-training pipelines for large language models.
• Improve model quality through supervised fine-tuning, continued pretraining, preference optimization, reinforcement learning, evaluation, and experimentation.
• Build and improve PyTorch-based training infrastructure, tooling, and developer workflows.
• Optimize distributed training across multi-GPU environments by improving throughput, memory efficiency, scalability, and GPU utilization.
• Investigate model training issues, including convergence, instability, communication overhead, and performance bottlenecks.
• Design evaluation methodologies, benchmark models, analyze failure modes, and acheive model improvements through experimentation.
• Collaborate directly with customers to understand real-world workloads and translate those learnings into improvements across Lightning AI's research platform.
• Partner closely with research, infrastructure, and platform engineering teams to build production-ready AI systems.
• Contribute to open-source projects through new features, tooling improvements, documentation, and community engagement

What You’ll Need

Required Qualifications

• Significant experience training, fine-tuning, evaluating, and/or optimizing transformer-based language models using PyTorch.
• Experience with modern LLM training and post-training techniques such as continued pretraining, SFT, RLHF, preference optimization (DPO, PPO, GRPO), reward modeling, or similar approaches.
• Strong understanding of distributed training and multi-node systems, with experience improving training performance, scalability, and/or efficiency.
• Strong software engineering fundamentals, including building production-quality Python software and research tooling.
• Experience designing experiments, evaluating model performance, and debugging complex training and/or optimization issues.
• Excellent communication and collaboration skills, including the ability to work effectively across research, product, infrastructure, and customer-facing engagements.
• Comfortable working in fast-moving, ambiguous environments where priorities evolve over time.
• Master's degree, PhD, or equivalent industry experience in Machine Learning, AI, Computer Science, or a related field

Ideal Experience

Experience with one or more of the following:

• DeepSpeed, FSDP, Megatron-LM, Hugging Face Transformers, TRL, PEFT, Lightning Fabric, NVIDIA Molt, or similar training frameworks.
• CUDA, Triton, vLLM, SGLang, TensorRT, or other AI systems and performance optimization technologies.
• GPU performance optimization, mixed precision, memory optimization, or distributed training optimization.
• Open-source contributions, research publications, or production AI platforms supporting large-scale training or inference workloads.
• Startup experience or experience working on highly cross-functional engineering teams.

Compensation