#thecompaniesapi early LLM results are bonkers but the qop/s are relatively slow; setting up results storage to train smaller models; one step at a time