Testing as an act of understanding.

WebQA Agent investigates how an agent can move through the web, assemble evidence, and answer questions whose truth is distributed across pages and interactions.

My contribution

As an early contributor, I worked across product design, frontend, and backend development—helping turn a research premise into a system people could operate and inspect.

Agent workflow
Evidence
Product design
Full-stack development

System architecture from the public WebQA Agent repository — Architecture from the public repository

Goal → action → evidence

“Confirm that a person can find a project—and understand what to do next.”

ObserveChooseAct & verifyPreserve evidence

Web questions are rarely answered by retrieval alone. The agent has to decide where to look, what counts as evidence, and when the evidence is sufficient.

Building the interface and the system together made evaluation concrete: every answer is also a trace of choices, uncertainty, and judgment.