This startup fine tuned their own LLM that performs as well as GPT-4o for production log files but 200X cheaper. Logs at scale are expensive so it turns an impossible cost into one that is quite viable.