NanoResearch

Why we included this project

Most AI research assistants stop at a literature summary or a code sketch, but NanoResearch is built to carry a research topic the rest of the way. It searches related work, plans an experiment, writes training code, and submits the job to a local GPU or a SLURM cluster, then parses the actual logs and turns the results into figures and a LaTeX paper. Because every table and chart comes from runs that really happened, the output stands up when someone checks it. Two details make it practical day to day: failed stages can be resumed instead of rerun from scratch, and you can route cheap models to boilerplate steps while saving the stronger ones for the harder parts. For researchers and small ML teams that want a paper draft grounded in real experimental output rather than generated numbers, this is a genuinely useful tool.