Problem

An environment with high turnover rate, like an e-Commerce fulfillment warehouse, can benefit from AI assistance to ensure the right orders are packed for delivery. However, as these warehouses typically house millions of products across many brands, it can be exhausting to manually manage the complexity of expanding product ontologies, while keeping up-to-speed with product turnover.

Solution

Layernext boosts a flexible API that integrates with third-party applications. In this case, it captures product data as soon as they are uploaded to individual brand inventories. Captured data is automatically curated on LayerNext, which triggers the creation of annotation projects that are sent to an off-site annotation team. Once the annotated data is inspected, they are automatically downloaded into LayerNext and fed to the model training process. These automated workflows allow the model to be trained and deployed at the highest speed of productivity, without much human intervention.

Result

By leaning on LayerNext’s automation features, the company created a viable AI model that is continuously trained and deployed, saving on the time and costs of building a similar product; and alternatively the cost of manpower to manage a fast-paced project of this scale.

A warehouse solutions company for logistics and fulfillment of e-Commerce products explores the use of computer vision to reduce fulfillment errors. Wrongly packed orders account for 30% of returns made on an annual basis, resulting in a chain of operating costs to rectify matters. To effectively train the computer vision model, the warehouse solutions company needed software that can manage the complexity of datasets containing large and expanding product SKUs and ontologies to keep up-to-speed with growing demands. Here’s how LayerNext met their needs.

Automated data collection using API

LayerNext boosts a flexible API that helps collect data from edge devices installed on customer sites. Datasets and metadata are fed into the LayerNext MetaLake by a third-party application, automatically curated and accessible as soon as they are added to individual brand inventories. Customers of the warehouse solutions company continue to manage their inventories as they always have, without being onboarded to a new platform.

Workflow automation in image annotation

Upon curation, the LayerNext platform automatically creates annotation projects to send datasets to an off-site team for object labeling. This ensures little to no project management, and the data engineering team only need to monitor annotated data.

Automated model training

Once datasets are annotated and inspected, they are automatically downloaded and fed into the model training process. By automating the data collection, annotation, and model training processes, the AI model is constantly trained and deployed at the highest speed of productivity.

The data engineering team monitors model accuracy using LayerNext Analytics, which provides both high-level visibility and in-depth data on model training. It also benchmarks production and training environments to ensure optimal model performance over time, which is crucial for complex, fast-growing datasets.

By utilizing LayerNext, the warehouse solutions company saved time and costs in building a similar platform themselves. Automated processes from data curation to model training saved them on manpower costs, while building a viable, productive computer vision model that worked for their business objectives.