Reading the label, automatically

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

The challenge had three focus areas, each one a step from a raw image towards a compliance decision.

A photograph of the product label is passed through optical character recognition, which reads every line of text on the packaging and returns it with a confidence score and its position on the label. On the sample labels this recognised the core fields, from the nutritional information table and reference values through to the full ingredient list and pack size, with high confidence.

The proof of concept met its goal. Working from real photographs of the client's products, the system read the labels, structured the content and demonstrated that generative AI could reason over the result for compliance, exactly the end-to-end path the engagement set out to validate. The approach was proven across several of the client's supplement lines.

• Accurate reading proven — The nutritional information, reference values, ingredients and pack details were lifted from label photographs with high recognition confidence.
• Structured output delivered — Extracted text was organised into the fields that matter for a compliance check rather than left as raw words.
• Compliance reasoning demonstrated — Generative AI on Amazon Bedrock showed it could reason over the structured label against another market's requirements.
• Handover — The client left the proof of concept with clear evidence that automated, image-to-compliance label validation is viable for its range.

With viability established, the natural next steps (projected) would be to widen coverage across the full product range, encode the labelling rules for each target market, and wrap the flow in a simple interface so compliance checks can be run routinely as new products and markets are added.