Key takeaway
- An AI platform use case: how Shelf Prism detects bottles, brands, counts and sizes from shelf images on AWS, with its architecture and what to check first.
An AI platform for product metadata uses computer vision to read shelf images and return which products are present, their brands, how many there are and what size they are. This use case follows Shelf Prism, an application for liquor outlets built with the Digitaiken team on AWS services.
It is written for retail, consumer goods and engineering leaders who want to automate stock visibility across many outlets. It covers the problem, the pipeline, the benefits, where AI agents fit and what to check first.
What is an AI platform for product metadata?
Shelf Prism is an application needed by liquor outlets in different locations, each holding bottles of different brands. The outlets needed to automate the detection of information about the liquor available in each outlet. XenonStack provided a solution to the Digitaiken team that uses AWS services for the app, covering model training, detection of liquor bottles, detection of available brands, the count of bottles and the size of bottles in a particular outlet.
What problem did the customer face?
Three problems sat behind the need. Data had to be collected from outlets in other localities, specific data on available brands was needed, and records of past data from every outlet were missing. The consequences were a lack of data on what was on the shelf, less availability of liquor, and a risk that customers visit less often when the liquor they want is not there.
How does the solution work?
The platform moves each shelf photo through five steps, from the app to the dashboard.
- The app captures images.
- An image check module tests orientation and other quality checks such as blur and low brightness.
- An object detection module detects bottles and boxes, then returns the number of objects, the brand and the size of each.
- The data is processed and stored in a database, with other AWS services supporting the flow.
- The required information appears in the app and on a dashboard.
The design has two architecture views, one for the machine learning pipeline and one for the platform. General object detection services work in a similar way: the Amazon Rekognition documentation describes detecting labels, objects, concepts and scenes in images.
What changes compared with manual outlet checks?
| Area | Before | After |
|---|---|---|
| Stock visibility | Staff reach each outlet to learn what liquor is present | Detection from images returns what is on the shelf |
| Bottle details | Counts and sizes depend on manual effort | Count, brand and size are detected automatically |
| Reporting | Records of past outlet data were missing | Results are stored and shown in the app and on a dashboard |
In plain terms, the manual effort of reaching every outlet falls away, and the business gets a stored record of what each outlet holds.
What benefits does this approach offer?
The page names three. Automated applications can combine inspection and quality control with AI. Inventory management improves because object detection can notify staff when merchandise needs refilling and let managers track product trends and place orders earlier. Automation also removes human bias, because people are limited by unconscious presumptions and by how much information they can process.

Where do AI agents fit in a product metadata platform?
Detection produces structured facts: which brand, how many, what size, at which outlet. An AI agent can act on those facts, for example by raising a restock request for a manager to approve when a count falls. This is a general pattern, not something the Shelf Prism build describes. ElixirClaw is XenonStack's AI agents computing platform for building and running such agents.
What should you check before adopting a similar platform?
- Decide which facts you need from each image, such as count, brand and size.
- Set image quality rules, so blurred or dark photos are rejected at capture.
- Agree where results are stored and who can see outlet level history.
- Define what should happen when detected stock is low, and who approves the action.
- Plan how the model is retrained as new brands and package sizes appear.
To map one decision your agents make about stock, and the data and policy it depends on, book a Context Workshop. See what happens in the workshop.
Frequently asked questions
What is Shelf Prism?
Shelf Prism is an application for liquor outlets that detects the bottles on a shelf, their brands, the count and the size of bottles from images, supported by AWS services.
How does the platform handle poor photos?
An image check module tests orientation and other quality issues such as blur and low brightness before the image reaches the object detection module.
What does the object detection module return?
It detects bottles and boxes and returns the number of objects in the image, the brand of each object and its size.
How does image detection help inventory management?
It can notify staff when merchandise needs refilling and let managers track product trends and place orders more efficiently.
Can AI agents use this data?
Yes, they can act on the structured results, such as a falling count of a brand at an outlet, provided the actions and approvals are governed.
Ready to put AI agents to work? See an ElixirClaw demo: the agentic OS to build, orchestrate and govern AI agents across your enterprise systems.
Explore the XenonStack platform family: ElixirData · ElixirClaw · ElixirHub · NexaStack · MetaSecure · Neural AI · AgentGRC · AgentSRE · AgentAnalyst · AgentInstruct · AgentRAI · AgentScrum · AgentSearch · AgentForce · AI Aviator · SAIF Aviator · CohortIQ · XenonStack AI · XenonStack UAE