Northeastern student builds AI agent that fixes chemical plants' blueprints

When Northeastern University industrial engineering student Sierre Ternoey started her research position in Aachen, Germany, she was challenged to fix an issue that frustrates chemical engineers worldwide, Northeastern Global News reported.

The industrial plants that make cleaning agents, clothing fabrics, and countless other products are designed with sophisticated software that lays out a "chemical flow sheet" — essentially a blueprint for how a chemical factory will work. The software simulates production processes to ensure ingredients are mixed and recycled while useful products are separated, but first drafts typically don't work perfectly.

"What you get back is a long, dense report telling you something broke without telling you why," Ternoey told Northeastern Global News. "Sorting out the cause takes real experience, a lot of time" and, in practice, a ton of guesswork.

Using skills she acquired in a first-year course at Northeastern, she produced an AI agent that could read the report, diagnose the problem, and fix the flow sheet to hand back a working simulation. "In the best configuration, the agent passed 26 of 30 test cases," she said, meaning it found a solution without disturbing the engineer's original design choices.

Her supervisor, research associate Jan Pyschik, described the project as "kind of a wildcard" that didn't quite fit his research agenda and was too uncertain to assign as a student's thesis. When other students considered it, they were advised to avoid it because "it wasn't clear if it was going to work at all."

Although she had never taken a chemical engineering course, Ternoey's Cornerstone of Engineering course at Northeastern gave her the tools — it "put me in front of problems before I had the tools to solve them and told me to start anyway." The course teaches the engineering design process, engineering ethics, computer-aided design, and programming, and partners with a local elementary school on open-ended exhibit projects.

The project shows AI agents moving into industrial engineering: automating the read-diagnose-repair loop that previously depended on experienced engineers is a natural application of generative AI inside professional software.

What to watch: whether the agent scales from 30 test cases to real chemical plants, and whether similar AI-repairs-the-simulation approaches can be replicated in pharma, energy, and other process industries.