Call for Client Projects — Winter 2027

AI Agents for Science (CMSC 35370) is seeking 4–6 research groups to serve as project clients this winter.

The deal

In the final weeks of the quarter (mid-February through demo day, March 10, 2027), teams of three students build a working AI agent for a real research group—yours—with instructor and TA support and course-provided compute. You get a working prototype, its complete execution traces, and a report; the students get a real problem with a real stakeholder.

Good client tasks are concrete, scoped to three weeks, and checkable. Examples of the right shape: Synthesize and organize the literature for a review in progress, with verified citations; triage and classify a backlog of instrument logs or pipeline failures; build an agent that monitors a data stream and drafts a weekly status report; automate a recurring analysis your group does by hand. If you are unsure whether a task fits, ask—scoping help is part of the offer.

What we ask of clients

A one-page task specification — the goal, what success looks like, and what is out of scope.
Working data and code access before the sprint starts — accounts created, data shareable, licenses checked. This is the most common point of failure, so we ask for it up front.
A named contact willing to spend about 30 minutes per week with the team during the three-week sprint.
Demo day attendance (Wednesday, March 10, 2027, 3:00–4:20pm) — you come see what they built.

Teams whose client project falls through revert to the course’s default project, so there is no penalty to students if circumstances change—but the four commitments above are what make a client project gradeable, and we hold to them.

How to propose

Send the one-page specification to Ian Foster (foster@anl.gov) by December 15, 2026. We will confirm selected projects by early January.