Ülemiste City Test City: Gala Sushi tests an AI-assisted checkout solution
How can restaurants make checkout faster and reduce manual work? Gala Sushi, located in Ülemiste City, is testing a camera-based checkout solution at its restaurant on Lõõtsa 4. The technology uses artificial intelligence to identify products and quantities selected from the buffet. By analysing camera images, the system suggests the items included in the purchase. At present, it covers both sushi pieces and drinks. Prices are retrieved from the checkout system’s product catalogue, and the proposed order can be reviewed and corrected by staff or the customer before payment.
The aim is to reduce the need for manual counting and product entry while making service smoother, particularly during busy periods. During the pilot, the system’s suggested orders are compared with verified results based on images of actual plates and trays. This makes it possible to assess how accurately the technology identifies different products and quantities, as well as how quickly it provides a result.
Development also focuses on more challenging situations, such as overlapping sushi pieces, products with similar appearances and reflections on drink packaging. To improve recognition, the team refines the instructions and compares different models and configurations using the same verified images. Corrections made by staff provide reliable examples for further development, but the system does not automatically learn from every transaction. The pilot will also measure overall order accuracy and potential time savings, meaning the solution’s performance cannot yet be expressed as a single universal percentage.
The Ülemiste City Test City environment allows Gala Sushi to develop the solution in a real restaurant setting. As the restaurant already operates in the campus, the technology can be tested during everyday service, with campus employees and visitors as its natural users. This helps reveal how the solution performs in real-life situations and what changes could make it more reliable. Although the system currently recognises sushi pieces and drinks, the underlying approach is not limited to these products. Adding other items requires expanding the product catalogue and validating recognition against new examples.
