Internships with real hardware.
Paid, project-based internships for RPI undergraduates and graduate students. Every track ends with something you can show someone — a technical report, a build guide, a preprint, a market study.
Four tracks.
GPU systems
Work directly on the Blackwell hardware. Benchmark training and inference runs, profile CUDA kernels with Nsight, and build tooling for the job queue.
You leave with a published technical report.
ML research
Work on the fine-tuning pipelines: implement an evaluation metric, run ablations on LoRA rank and alpha, or benchmark a model family against our NTQS harness.
You leave with co-authorship on the write-up.
Infrastructure
Build and document a reference AI server end to end: spec the parts, assemble the machine, install the OS and CUDA stack, run the burn-in suite, publish the guide.
You leave with a hardware build guide with your name on it.
Product & outreach
Run the campus go-to-market: poster campaigns, the Discord, info sessions with clubs, and interviews with student researchers about what they actually need from compute.
You leave with a market report of your own.
Not that much, honestly.
- Curiosity over credentials. Whether you want to understand the machine matters more than whether you have used one before.
- You finish the thing. Every track ends in something real, and the point of the placement is getting there.
- You say when you are stuck. “I have been going in circles for two days” is useful information, not an admission. We would much rather know.
Apply for a track.
Tell us your year, your major, which track interests you, and one thing you have built or taken apart.