Evaluating Pallet’s memory layer.
Freight operations require client-specific rules and artifacts to make accurate decisions. Memory-Bench measures how our memory layer serves that knowledge to Pallet’s AI agents — lifting first-try accuracy from 22% to 62% across seven frontier models.
Memory-Bench: what an organization’s memory is worth to an AI agent
Pallet’s memory layer moves an organization’s tacit operational knowledge into the platform. Memory-Bench measures what that memory is worth to Pallet’s agents — accuracy, reliability, and cost across seven large language models, with and without it.
Cost-to-serve: the buy rate, taken apart
For a brokered load the buy rate is the largest piece of the cost. It reads as one price but is really three: marginal cost, opportunity cost, and carrier rent. We take the split apart and show how sensitive each part is to fuel, volume, and competition.
The Hard Parts of Fine-Tuning Our First Production Model
How we fine-tuned and deployed a focused open model with frontier-level accuracy at roughly one-third the cost.
Introducing Pallet Labs
Turning frontier-model capability into AI that reliably participates in real supply chain operations takes research. Pallet Labs is our dedicated effort to study the hardest and most important problems in applying AI to one of the world’s most complex industries.
