Inside the Research Blog: Post-Training Experiments That Prove the Data Works

AfterQuery's research team publishes detailed post-training runs: a 5x Terminal-Bench gain with Tinker and Harbor, on-policy distillation on GDPval, PivotRL on banking tasks, GRPO on SWE tasks, and a Nemotron credit model with Hebbia. This episode distills the recurring technical lessons so engineering and research candidates can discuss them intelligently.