Apply for a GPU community grant: Personal project

#1
by ParetoOptimal - opened

OffGridSchedula is my solo side project: a local-first agent that turns a messy class/group chat (or a flyer screenshot) into reviewed calendar events, conflict checks, and a drafted reply, all from a phone browser. It runs my own QLoRA fine-tune of Gemma-4 E4B (~4B params, published on the Hub) through llama.cpp with vision, no cloud AI APIs. I publish redacted run traces as a public dataset and the full eval-gated training pipeline. I'm requesting a T4-small grant to keep the live demo running so others can try a real small-model agent and learn from the open model + traces.

Sign up or log in to comment