Warehouse intelligence, in development

See your warehouse. Find its next move.

RackSort is being built to turn warehouse images, inventory, and demand into better decisions about where stock goes and what happens next.

Explore a warehouse Explore the model

Built on SmallDecide. Designed to work with ERP.AI.

A warehouse, in viewIllustrative example

An incoming pallet needs a homeFind a suitable open bin.

Combine observed space with item compatibility, capacity, and current reservations.

A scripted illustration of the planned workflow. No live vision inference or stock changes.

The direction

Less searching.
More purposeful movement.

Connect what is visible on the floor with the records and constraints that make a warehouse work.

Give incoming stock a place to go.

Match available space with the item's dimensions, storage requirements, and demand. Keep uncertain observations visible for review.

Keep the next pick within reach.

Bring stock from reserve into picking locations as demand changes, accounting for work and incoming moves already planned.

Make the space work harder.

Identify stock-placement opportunities and compare the benefit of a move with the effort it takes to make it.

How it is designed to work

A clear path from observation to operation.

RackSort proposes the next move. ERP.AI can give your team the records, review steps, and tasks to carry it out.

Understand the floor

Connect warehouse images with known bins, stock identities, dimensions, and order demand.

ImagesInventoryDemand

Find a useful move

Compare feasible options, then check the proposed operation against warehouse rules and its supporting evidence.

ConstraintsSmallDecide

Complete it in ERP.AI

Review a recommendation, assign the work, and confirm the actual movement with scans and an auditable record.

TasksScansConfirmation

Vision, optimization, and the ERP.AI application connection are planned components.

A decision the engine makes today

A lost reply should not buy a second label.

A label request reaches the carrier, but its reply never arrives. Retrying with a new request could pay for the same shipment twice.

RackSort returns reconcile: establish what happened before another billable operation. Each verdict includes a next step, supporting facts, and the specification clauses it applies.

Read the full model overview

Carrier requestWorked example

{
  "decision": "carrier",
  "verdict": "reconcile",
  "state": "Uncertain",
  "next": {
    "action": "lookup",
    "text": "Look up request CR-0412 with the carrier and record what it finds before anything billable runs again."
  }
}

Generated from the rule-based engine. A verified idempotent connector permits retrying the same key; finding an existing label leads to replay.

Five verdicts, each with a next step

allow: Proceed
replay: Reuse the result
reject: Stop the operation
review: A person decides
reconcile: Establish the facts

Explore all ten decision types

The work so far

A working foundation.
A clear next step.

Explore the implemented decision engine, its examples, and the results behind the model.

Explore all ten decisions

Implemented · experimental

Warehouse decision engine

Rules and a fine-tuned SmallDecide model review receiving, reservations, counts, packing, carrier actions, and more.

7 of 11 questions qualify for model 47ed8d6e03938bfc (float32-cpu) at a 3% error budget. On 6,598 held-out rows the model settles 947 verdicts that the rules alone send to review or reconciliation, with 1 unsafe verdict. Evaluations has every number.

Planned

Warehouse vision and optimization

Image-based observations, a model of the current warehouse, and a planner for put-away, replenishment, and stock placement.