On-Device AI
Models that ship inside the product — no server, no round-trip, no excuses.
eSPUD is an independent research collective exploring the frontier of on-device AI: models small enough to run on phones, sensors, wearables, and the constrained hardware that powers everyday utilitarian devices.
We believe meaningful intelligence shouldn't require a datacenter round-trip. Our work spans efficient architectures, compression, hardware-aware training, and the systems plumbing required to ship AI that respects latency, energy, and the user's privacy.
We take AI ethics seriously. Every project is reviewed for data-use, consent, bias, and downstream impact before it leaves the lab — and we publish what we learn, including what didn't work.
Models that ship inside the product — no server, no round-trip, no excuses.
Pruning, quantization, distillation — getting more from fewer parameters and watts.
The runtimes, compilers, and orchestration that make tiny models behave at scale.
Partnering with industry to put ideas in front of real users on real devices.
A research hub for the Principal Investigator.
pi-hub is a local-first, installable web app that consolidates projects, team coordination, and academic deadlines into a single interface — built on the principle that nothing you type leaves your browser. No backend, no accounts, no telemetry. Just a fast, private notebook for running a research group.