My current research focuses on RAG systems and methods that help language models use external knowledge more reliably.
Areas of interest include grounded generation, small language models, in-context transfer, evidence-use evaluation, and practical architectures that reduce costs and improve controllability.