Teaching small models to read job postings: our arena and the JOA Small, Advance and Deep fine-tunes

Our job-posting extraction arena, and how a fine-tuned 3B model reached 1st of 84 (benchmark dated 2 Oct 2026).