Until yesterday, artificial intelligence on the factory floor was a silent bystander: analyzing data, generating graphs and suggesting optimizations.
Today, artificial intelligence is not just observing, but acting.
The Zerynth Case
Zerynth is an Italian deep-tech company developing an IIoT platform for the rapid and scalable digitization of industrial manufacturing. It is currently evolving its solution into an “Industrial AI Copilot” platform equipped with a conversational agent for natural interaction with data and production processes. The development of this artificial intelligence component is considered strategic for the company’s product evolution and future positioning.
The problem: the technical barrier between man and machine
Configuring an industrial monitoring system or changing production parameters often requires specific technical skills and time spent among complex software. This “barrier to entry” slows down operations and increases the risk of manual errors.
The challenge of our experimentation was clear: to make the factory not only responsive, but self-configurable through natural language.
The Solution: AI agents with implementation capabilities
For this project, we developed a modular architecture based on Agentic AI. Unlike a standard chatbot, this system has “digital hands” (via API and MCP protocols) that allow it to interact directly with the Zerynth platform.
Here are the three technological pillars of the solution:
- Agent Orchestration: Using the Agno framework to manage agents specialized in different tasks: from cost analysis to device configuration.
- Cloud-to-Edge Integration: Using Zerynth’s Jobs API, textual input is translated into a physical command sent directly to the IoT gateway at the machine.
- Rules Engine Automation: The AI is able to independently write alarming and automation rules, translating an operator’s wish into correct technical syntax.
Safety First: Human-in-the-Loop
Interacting with heavy machinery requires extreme caution. Therefore, the system integrates AI Safety and Explainable AI (XAI) logics. The agent never performs a critical command autonomously: the system precomputes the action and always requires explicit confirmation from the user (Human-in-the-loop). In addition, each action is accompanied by an explanation of why the system is suggesting it, dramatically reducing operational risk.
What we learned from the tests
Validation in the demo environments and at the MADE competence center showed exciting results, but also highlighted the challenges of the future:
- Context accuracy: The AI excels at understanding temporal queries (“this week”) and complex filters (“only A-line presses”).
- Digitization of manuals: We have found that handling image-rich manuals requires advanced pre-processing so that critical details are not lost during conversion to text for the RAG system.
- Name mapping: It is critical that the AI knows that “the big press” corresponds to the technical ID “PR-094,” a semantic mapping task we are continuing to invest in.
Takeaways for the future of the industry
The success of this phase of the AI-MATTERS project opens the door to a factory where:
- Efficiency is in your voice: Less time lost in configurations, more time for strategy.
- Technology is inclusive: Even less software-savvy personnel can handle complex configurations safely.
- Interaction is proactive: The system does not wait for orders, but suggests parameter changes based on energy costs or production loads.
The collaboration between AI-MATTERS and Zerynth is proving that conversational control is the key to democratic and secure digitization. The factory of the future is not commanded by codes, but by dialogue.