Designed / Selected work
RAG and voice agents
Answers with context, and conversations that feel more natural.
From a question to a useful answer.
People need answers grounded in relevant material and voice interactions that feel responsive.
What it does
Find context
Bring relevant information into view.
Stay grounded
Keep answers connected to the question.
Feel natural
Make room for a more fluid conversation.
Outcomes & scale
Source and caveat
Historical 40-question unseen retrieval comparison, 18 November 2026
CV rounds 43.1% versus 36.0%. Current 38-question set: 42.3% versus 34.4% (+7.9 points) after two questions about a private topic were withdrawn; retrieval recall is not answer correctness
Source and caveat
CV-reported description; see the linked CV for scope.
A descriptive claim, not a measured success rate unless stated.
Source and caveat
CV-reported description; see the linked CV for scope.
A descriptive claim, not a measured success rate unless stated.
Scope
Designed. The CV does not report a public answer-quality score or speech latency for these agents.