Capital One AI Research

Focuses on the development of large-scale foundation models across various data modalities. Includes model types like Language, Graphs and Time Series, pretraining, RAG (Retrieval-Augmented Generation), and fine-tuning strategies and how to integrate Knowledge Systems and Multimodal Knowledge Representation to enhance factual understanding.

Dedicated to building next-generation AI systems that act and reason independently and collaboratively. Includes single and multi-agentic systems, advanced complex reasoning techniques, resource management models like Test Time Compute and Mixture of Experts, and the use of Reinforcement Learning (including RLHF and World Models) to train autonomous behaviors.

Research geared toward ensuring AI systems are safe and reliable, especially relevant in our highly-regulated industry. Core areas include LLM Safety and Security, defining benchmarks and evaluation frameworks, Privacy Preservation, Unlearning (to remove specific data influence) and reliable decision-making through Causal Inference.

Centers on solving the engineering challenges of operating models at scale. Involves improving the entire pipeline, including LLM Training, Inference and Optimization techniques, to enhance the speed and resource efficiency of deployed systems.

Includes rigorous data curation, generation of reliable synthetic data, and the application of models to understand and predict real-world dynamics, such as User Behavior Modeling.