Recognising every bottle on the shelf

A Retail business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had three focus areas, each anchored in making image recognition dependable enough to build a product on.

The engagement followed Cloud Combinator's Cloud Accelerator Programme, moving deliberately from understanding the product to delivering a working solution in the client's account, with upskilling built in along the way.

The proof of concept delivered a working, serverless image-matching pipeline in the client's own AWS account and demonstrated Amazon Rekognition as a suitable machine learning environment for continuing to develop the client model. Acceptance was defined by signing off

• Working pipeline — A serverless image-matching pipeline built on Amazon S3, API Gateway, Lambda and Rekognition, deployed in the client's own AWS account.
• Real-world testing — The model was exercised against a diverse dataset of high and low quality bottle images to establish genuine capability and limitations.
• Accuracy benchmark — A 60% accuracy rate was set as the initial target for real-world identification, providing a measurable bar for the proof of concept to validate.
• Upskilling — An immersion day and hands-on labs left the client's technical lead able to continue developing the model independently.
• Handover — Best-practice foundations for security, billing and recovery, along with a Well-Architected review, were handed over with the pipeline.

The delivered scope, confirming the technical competence of the client's internal lead, and confirming Rekognition's fit for ongoing development.

With a proven pipeline and an upskilled team, the client is positioned to refine accuracy, extend recognition across a wider catalogue and integrate identification into its customer-facing products as the next step.