Rocket science just got an AI copilot

What started as a faster way to build a rocket resulted in Peter Pak reimagining the way engineers build.

The Carnegie Mellon University mechanical engineering Ph.D. candidate developed RocketSmith, an AI-powered engineering system that integrates design software, flight simulations, and additive manufacturing into a single workflow. By combining large language models with traditional engineering tools, Pak alongside Amir Barati Farimani, professor of mechanical engineering, reduced a process that typically takes months to just days.

Historically, designing a high-powered rocket required engineers to move between various specialized software. One program creates the flight map, another generates CAD models, and others prepare parts for manufacturing. Each of these iterations can take days or weeks, making experimentation slow. RocketSmith streamlines this process. After users describe what they want their rocket to accomplish, the AI coordinates multiple engineering tools, including OpenRocket, to generate the design, run flight simulations, create 3D-printable CAD models, and even recommend improvements before the rocket is built.

“It’s not just generating the design,” said Pak. “The system gives you feedback on how well that design will perform. It predicts when the rocket will reach its highest altitude, when the parachute should deploy, and how different design choices affect the flight. It allows us to make changes much faster.”

The project reflects a broader shift in engineering, where AI is becoming a collaborative design partner rather than just a tool.

“RocketSmith does not eliminate the need for engineering knowledge. We still need to understand rocket stability, recovery systems, safety requirements and how to interpret simulations,” Pak said. “There is a lot of engineering knowledge that only comes from experience, so even when AI helps us to accelerate the workflow, we still need the domain knowledge to make good engineering decisions.”

To evaluate the system, Pak and the research team designed, manufactured, and launched four high-powered rockets. While each rocket successfully launched, only two returned in re-flyable condition.

“Recovery is actually one of the hardest parts of rocketry,” said Pak. “A lot of rockets end in failure, but each launch teaches us something that makes our next designs better.”

The cycle of prototyping, building, launching, and starting over is what drew Pak to rocketry. As a member of Carnegie Mellon’s Rocket Command, a student rocketry organization, Pak is part of the team of students that competes in NASA’s Student Launch competition, a research challenge that provides cost-effective research and development to support the Space Launch System and Artemis missions.

Our students aren’t just learning how to use AI, they’re learning how to build the next generation of engineering tools.

Amir Barati Farimani, Professor, Mechanical Engineering

“Our students aren’t just learning how to use AI, they’re learning how to build the next generation of engineering tools,” said Barati Farimani. “By combining engineering fundamentals with AI, they’re building technologies to accelerate many fields like aerospace and manufacturing.”

Pak believes that future versions of RocketSmith could incorporate advanced analyses including computational fluid dynamics and other high-fidelity simulations to allow engineers to evaluate even more complex designs before they are built.

“The technology will keep changing,” said Pak. “But engineering is still about solving problems, learning from failure, and building something better next time. AI gives us another tool to do that.”