Director, Generative AI Platform & Training

Capital One; San Francisco (CA); On-site; USD 180,000 - 230,000; Full time

Job summary

Capital One is building responsible and reliable AI systems to transform banking. The Generative AI Training team runs GPU clusters and distributed training for foundation models at scale, enabling experiments with KServe and vLLM.

You will help shape the enterprise AI strategy and guide talent development. You will partner with engineers, scientists, and PMs to deploy, test, and operate core AI systems, balancing innovation with governance and risk considerations.

Qualifications

Responsibilities

• Partner with cross-functional team of engineers, researchers, and PMs to deliver AI-powered products.
• Oversee design, development, testing, deployment, and operation of the platform's core systems: distributed training and fine-tuning, RL workflows, GPU scheduling, resiliency, and self‑service environments.
• Make build-vs-buy decisions across Open Source and SaaS AI tech such as AWS Ultraclusters, HuggingFace, VectorDBs, PyTorch.
• Invent state-of-the-art techniques to improve scalability, cost, throughput, and reliability of large-scale distributed training.
• Own GPU capacity planning and cost governance across training workloads.
• Contribute to long-term technical vision and roadmap of foundational AI systems.
• Attract and retain top AI talent and foster a culture of learning and responsible AI.

Skills

Distributed systems

GPU training

MLOps

Open Source

Cloud platforms

Tools

PyTorch

HuggingFace

KServe

VectorDBs

AWS

Job description

Capital One is building responsible and reliable AI systems to transform banking. The Generative AI Training team runs GPU clusters and distributed training for foundation models at scale, enabling experiments with KServe and vLLM.

You will help shape the enterprise AI strategy and guide talent development. You will partner with engineers, scientists, and PMs to deploy, test, and operate core AI systems, balancing innovation with governance and risk considerations.