DealMind

DealMind: negotiators that learn from every deal. A team of AI negotiator agents buys furniture from suppliers for a shop chain, one specialist per supplier, running on Nemotron 3 Nano via Crusoe Managed Inference. After every negotiation, each negotiator writes a report. A coach agent (Nemotron 3 Super) reads them, maintains a shared playbook with a dossier on each supplier plus general tactics, and changes one strategy knob at a time, keeping a change only if the score measurably improves. Supplier behavior and scoring are enforced in code, so results can’t be faked by a model. In 5 rounds, every supplier’s deal score improves: Northwood goes from a walk-away (0) to 77. On a supplier the team has never met, a negotiator briefed by the playbook beats a blank one, 82.5 vs 67.9. Optional NVIDIA Switchyard routing escalates negotiators from Nano to Super.