Viso.ai is a major coffee retailer with around 300+ Coffee Retail Shops within Switzerland. Viso.ai wanted to detect & determine empty/partially empty/foreign objects placed within compartments from the condiment station within an image which was streamed from the Coffee Shops CCTV Camera’s every 5 mins.
Noventiq implemented an Agentic AI Solution wherein the following Agents Were Deployed using Amazon Bedrock Agent Core Functionality.
- Image Analysis Agent – QWEN 2.5 7B Vision Language Model was deployed as a custom model within the Bedrock Import Custom Model Functionality. This model was responsible for identifying the bounding boxes of empty and partially empty compartments. In the first pass, Bedrock generates prompts to identify areas of interest (e.g., an empty compartment within the condiment station). In the second pass, it produces OpenCV-compatible b_box2d coordinates (x1,y1,x2,y2) for the detected regions.
- Notification Agent – The Notification Agent was responsible of notifying the end user through dynamically generated image with bounding boxes overlayed and tagged (From the previous b_box2d coordinates) to identify areas of interest and the same was displayed back to the Coffee Shop Attendants/Baritas through whatsapp channel.
Customer Name
Viso.ai GmbH
AWS Product/Service
Amazon Bedrock, Lambda, S3, CloudFront
Additional Information
This is a unique use case in which the open-source QWEN 2.5 7B Vision Language Model was deployed in Amazon Bedrock using the Import Model feature, driven by the customer’s requirement for image analysis and inference capabilities on a pay-as-you-go GPU model.