Every freight forwarder faces two expensive questions. First: "When should we book the next container to India?" Book too early and you ship air. Book too late and packages pile up in the warehouse for weeks. Second: "200 packages, one container; how do you fit them?" A wrong answer means wasted space, crushed goods, or a second container at $2,000 or more.

Before Voxarel, the answer to both was gut feeling. A warehouse manager would eyeball the pile, call the shipping line, and hope for the best.

The Container Fill Prediction Engine

The prediction engine analyzes 90 days of historical CBM (cubic meter) accumulation data. It factors in seasonal patterns: Ramadan and Eid drive shipment surges on the UAE-India corridor. Diwali creates a secondary peak. Summer months see reduced volumes.

The engine calculates a confidence score for each prediction. When confidence exceeds 85%, the system generates an automatic booking recommendation. Below that threshold, it surfaces the prediction with supporting data for the operations manager to decide.

A feedback loop tracks actual versus predicted outcomes. Every container that ships updates the model's accuracy metrics. Over time, predictions improve because the model learns each tenant's specific booking cadence.

3D Bin Packing

The spatial problem: a container has fixed dimensions. The packages have variable dimensions, weights, and handling constraints. Fragile items sit on top. Liquids cannot be rotated. Heavy items go on the bottom for stability.

Voxarel integrates with the SeaRates API for 3D bin packing. The system sends all package dimensions, weights, and constraints. The API returns (X, Y, Z) coordinates for each package's optimal position. The result renders as an interactive 3D visualization built with React Three Fiber. The warehouse team can rotate the view, zoom in, and see exactly how to load each package before they start.

A PDF export provides a step-by-step loading sequence for staff who prefer a printed reference.

The Fallback Algorithm

External APIs need fallbacks. The SeaRates API has monthly request limits and occasional downtime. When unavailable, Voxarel falls back to a local wall-building algorithm. It sorts packages by height (tallest first), builds "walls" from floor to ceiling, and fills gaps with smaller items.

The wall-building approach produces less optimal packing, but it produces a valid loading plan immediately. The operations manager sees a flag indicating which algorithm produced the result.

The Break-Bulk Decision

Sometimes one container is not enough. The system calculates whether total CBM exceeds capacity. If it does, it evaluates two options: one larger container (20ft to 40ft upgrade) or two smaller containers (break-bulk). The cost comparison includes the container rate, surcharges for oversized items, and deadline pressure.

This decision used to happen over a phone call. Now it happens automatically with cost data attached.

The Numbers

A freight forwarder booking 4 to 6 containers per month saves $2,000 or more per avoided wrong-sized booking. Over a year: $24,000 to $48,000 in prevented waste. The packing optimization adds further savings by maximizing utilization and reducing break-bulk frequency.

The Pattern

AI in industrial software solves operational problems. The prediction engine reads 90 days of CBM data and recommends when to book a container. The bin packing solver calculates spatial coordinates from physical dimensions. No hallucination. No generated content. Find the expensive decision currently made on gut feeling. Build a model using data the company already collects. Provide confidence scoring so humans can override. Track accuracy over time.