Successful adoption of artificial intelligence is not about a single “magic” model. According to Anton Gertli, a senior solution architect at Red Hat, it rests on three pillars: people, processes, and technologies. In a practical retail demonstration, he showed how important it is to choose the right model, automate the entire cycle, and handle it from data all the way to deployment at the edge.
AI lifecycle: from goals through data to monitoring
An AI project must start with a clear definition of business goals—not with an attempt to deploy AI “because it’s trendy.” Next comes work with data: collection, storage, cleaning, and preparation, without which an accurate model will not emerge. Only then comes model development, its integration into applications, and ongoing monitoring of performance, accuracy, and deviations; the entire cycle is ongoing and never ends. Since it involves the business, data scientists, developers, and operations, a unified platform helps—Red Hat OpenShift (including OpenShift AI)—which brings together development, deployment, and automation, and thanks to its openness minimizes vendor lock‑in, whether it runs on‑prem or in the cloud.
Retail example: small models, big impact
In a “smart store” demonstration, participants photographed Horalky bars so the system could recognize the product and display information or price. The solution operated in two locations: at the edge (edge) for millisecond response times and in the data center for training new versions of the model. Key finding: a small predictive model of roughly 2.8 MB was sufficient for recognition, without a GPU and with training on the order of minutes—exactly according to the needs of the use case. An important component is also secure transfer between locations and a data pipeline that automates collection, training, validation, and deployment.
Gertli emphasized that “bigger” doesn’t always mean “better”—not every use case needs an LLM or cloud resources. Model hallucinations are addressed through tuning and methodologies, for example the “teacher model” approach (mentioned with IBM Granite models), but data quality and cleaning remain decisive. Questions from the audience confirmed that the platform can also be deployed fully on‑prem, which is important for regulated or security‑sensitive environments. And what about jobs? The technology is more likely to change the nature of work and create new roles than simply replace people.
