Research
AI Agent Systems
Language-model agents that plan, call tools and act over real data, and the evaluation needed to trust them.
An agent that can query a database, call an API and write back is far more useful than a chatbot — and far more dangerous when it is wrong. The research questions are about control: how an agent decomposes a task, how it decides which tool to call, how its intermediate steps are verified, and how failures are contained.
We are interested in agents grounded in data systems: planning over schemas, verifying generated queries and code before execution, cost-aware tool use, and benchmarks that measure the outcome of a task rather than the plausibility of the transcript.