I work at the intersection of AI, mathematics and building things that ship — instruments and tools for language models running on hardware I control, and native apps that keep the model on the device. Northeastern University, BS Mathematics and Business Administration (Fintech concentration), class of 2028. Looking for an AI/ML co-op from January 2027.
The through-line in everything below is one rule: a measurement is a number;
the absence of one is —. Most of the code in these repos exists because some
model or some API reported a plausible wrong value and something had to refuse it.
| mac-ai-activity | Live instrument panel for local AI inference. Joules per token, per-slot llama.cpp telemetry, SMC die temps and fans. Observe-only, native SwiftUI, no dependencies. |
| llmctl | Which local models are holding your Mac's RAM, and one command to get it back — including the idle server processes still pinning weights with mlock. |
| wake-on-llm | Turn a big local model on and off from your phone, from anywhere, over Tailscale. The last-resort path is a literal Wake-on-LAN packet. |
| solvr-ai | Native iOS and Mac Catalyst maths tutor: photograph a problem, get the working and the lesson behind it. Runs entirely on Apple's on-device model — no API key, no server, no network. Built and working, not released. |
| circadia | Hour-by-hour jet-lag plan for iPhone and Mac, simulated forward from the published phase-response literature: when to get light, when to avoid it, when to sleep. On device. Built and working, not released. |
| ascend | Gamified goal tracker for iPhone. Apple Intelligence breaks an ambition into milestones and daily steps on device, and the roadmaps are scored by a harness rather than asserted to be good. Built and working, not released. |
Solvr came out of my own tutoring practice — I have taught calculus and statistics since 2022 and kept rebuilding the same lesson plans by hand.