From Raw Data to Insights in Seconds: The AI Reporting Showcase

Predictive maintenance and condition monitoring are on everyone’s lips in modern industry, but the reality on the factory floor often looks quite different. Imagine a shift supervisor preparing for the next shop floor meeting and urgently needing a visualization of the utilization data for a specific production unit.

In the past, this process was often blocked by system gaps and delays. The problem is usually the complexity of the existing infrastructure. Data resides in isolated IoT backends, access requires special permissions in BI tools, and manually consolidating various data sources is so time-consuming that information is often available too late. If data first has to be laboriously exported, cleaned, and standardized, data-driven decision-making remains a theoretical ideal.

With our AI Reporting Showcase, we resolve this bottleneck. We combine the flexibility of AI agents with industrial machine data to create interactive reports directly within the interface, without any technical expertise or manual preparation.

The foundation: The democratization of machine data

The adesso Smart Product Platform forms the basis for our showcase. This is where data from assets (machines, sensors, etc.) converges. While our current showcase still uses demo data and simple CSV imports to enable a straightforward demonstration, the roadmap is clearly defined: A seamless connection via REST API will enable direct access to live data in the future.

Our goal is the democratization of data. Information must be available where it is needed for decision-making. Through generative AI, we bridge the gap between complex database structures and intuitive natural language. We have already presented a similar showcase for interacting with data from the adesso Smart Product Platform via chat in the blog article on the “Talking Machine”.

The Added Value of Dynamic Interactive Charts

A core problem with conventional AI approaches is their static nature. A bot often delivers only text or a static image. This is precisely where our showcase with dynamic interactive charts comes in. Our agent system does not deliver a finished image, but rather an interactive data object that is displayed as a chart in the frontend. This approach opens up various possibilities for users:

• True interaction: Users can zoom into the diagrams and closely inspect individual data points via mouseover.
• Dynamic expansion: If the user realizes, “I also need a temperature curve,” the existing diagram is adjusted on the fly.
• Flexibility: A simple command like “Show me this as a bar chart instead” immediately changes the display. The hurdle of having to manually configure visualization parameters is completely eliminated.

Behind the Scenes: The Agent Framework

To achieve this flexibility, we rely on agent-based orchestration. Instead of rigid program code, a team of specialized agents processes the user’s request:

• Planning Agent (Architect): It analyzes the user’s intent. If the request is unclear, it asks specific follow-up questions. It uses knowledge of the data schema to create a plan (calculations, chart type).
• Logic Agents: These agents access specialized functions. For example, they calculate the duration of status states (e.g., Idle vs. Active) or determine meter reading deltas over specific time periods.
• Visualization Agent: It creates the specific configuration for the visualization’s data object. To do this, it uses predefined functions that only need to be called with the correct parameters to generate an appropriate chart.
• Response Agent: It summarizes the technical results and formulates the response in natural language for the user.

This modular structure ensures that the system understands the context of the entire conversation.

Practical Example: Milling Machine

To make the theory tangible, let’s take a look at a demo milling machine. Here, we see how the interaction of the agents makes complex data usable on the fly.

Scenario 1: A technician asks: “How long was the machine in Operation Mode setup? Compare all categories.”

First, the planning agent creates a strategy. The logic agent handles the calculations and precisely sums the duration for each status category. Since the request is for a comparison, the system automatically opts for a visualization. The calculated values are displayed precisely in their SI unit (seconds) (see Fig. 1). The user thus receives not only a number, but also immediate context regarding the remaining runtime.

Scenario 2: In the second example, we compare temperature, vibration, and energy consumption.

After a plan has been created, the logic agent calculates the average values and the visualization agent presents them as a bar chart (Fig. 2). However, since the time series is often crucial for a deeper analysis, a simple command is enough to change the chart type. The visualization agent converts the view into a line chart (Fig. 3).

This agent-based approach allows for effortless adaptation to company-specific preferences during individual implementation. Whether it’s specific chart types for certain data sources or customized KPI logic, reporting adapts to the user, not the other way around.

Conclusion: Reporting as a dialogue, not a ticket

The showcase proves that the gap between the IT department and the shop floor can be bridged. For IT management, this means fewer ad-hoc requests for simple reports and greater data acceptance within the company. For technicians, it means real decision-making power through insights in seconds.

What’s the situation like in your company? Do your experts spend more time in BI tools or Excel than at the machine? Let’s discuss how we can mobilize machine data. Contact us for a demo and further insights into our AI Reporting Showcase.