Manufacturing SMEs, particularly in sectors with intensive logistics and traceability requirements such as food production, still rely heavily on manual data entry and screen-based interfaces for warehouse and inventory operations. Operators frequently need to interrupt physical tasks to input information into terminals, scan codes, or navigate complex software menus. This constant switching between physical handling and digital interaction increases execution time, creates opportunities for human error, and adds cognitive strain to workers.

In the case of Conservas Pinhais & Cia, logistics processes such as goods reception, picking preparation, stock movements and inventory counting require precise data registration to ensure traceability, compliance with food safety standards, and accurate stock control. Even small registration errors can lead to discrepancies in inventory levels, delays in order fulfilment, and additional verification procedures. In regulated environments such as the agri-food sector, maintaining accuracy and auditability is essential, making reliable data capture a critical operational requirement.

Another significant challenge relates to workforce onboarding and training. Traditional Manufacturing Execution Systems often require structured training before operators become fully autonomous. This creates a dependency on experienced staff and extends the learning curve for new employees. In a context where SMEs face increasing labour shortages and the need for higher operational agility, simplifying interaction with digital systems becomes a strategic priority.

From a technological perspective, integrating conversational artificial intelligence into industrial systems also presents challenges. Voice recognition must operate reliably in noisy factory environments, system integration must be secure and interoperable with existing software, and data handling must comply with European data protection regulations. Ensuring transparency, user control, and GDPR-compliant data management is fundamental when introducing AI-based assistance into operational workflows.

The CALIE experiment addresses these combined operational, organisational, and technological challenges by rethinking how workers interact with digital systems in logistics and inventory management, moving from screen-based input to natural, voice-enabled interaction while preserving traceability, compliance, and system reliability.

Objective 1:

Design and develop a voice-enabled Digital Intelligent Assistant integrated with the Flow Manufacturing system, enabling operators to execute logistics and inventory tasks through natural language interaction in a secure, modular and interoperable architecture aligned with the WASABI framework.

Objective 2:

Deploy and validate the assistant in real factory operations at Conservas Pinhais & Cia, demonstrating measurable improvements in logistics performance, including reduction of human errors, faster task execution, increased productivity, reduced training effort and improved operator satisfaction.

Objective 3:

Package and publish the developed conversational skill in the WASABI White Label Shop, ensuring replicability, scalability and accessibility for other manufacturing SMEs interested in adopting human-centric AI solutions for logistics and inventory management.

The CALIE experiment is implemented in the agri-food manufacturing sector, specifically in canned fish production. Conservas Pinhais & Cia operates a traditional yet export-oriented manufacturing facility in Portugal, where logistics, traceability, and inventory control are critical to daily operations. The experiment will be carried out directly at the company’s production and warehouse site, in a real operational environment rather than a laboratory setting.

The assistant will be deployed within the existing logistics and inventory workflows supported by the Flow Manufacturing system. The operational environment includes goods reception areas, storage facilities, picking preparation zones, and inventory control processes. These activities involve frequent material handling, barcode scanning, stock confirmations, and data registration tasks that must comply with strict traceability and food safety requirements.

The primary users of the solution are warehouse operators, logistics staff, and supervisors responsible for stock accuracy and order preparation. The system must therefore operate reliably in a factory environment characterised by background noise, movement, and time-sensitive operations. Usability, clarity of voice interaction, and minimal disruption to existing workflows are key considerations.

The experiment must also comply with regulatory and operational constraints. As a food manufacturing company, Conservas Pinhais & Cia follows strict food safety and traceability standards, requiring accurate and auditable data registration. In addition, the solution must ensure secure authentication, controlled access to operational data, and full compliance with European data protection regulations. Integration with the existing Flow Manufacturing system must preserve data integrity and ensure real-time synchronisation of inventory records.

By deploying the conversational assistant in a live manufacturing setting, the experiment demonstrates how voice-enabled interaction can be embedded into real industrial workflows without compromising compliance, traceability, or operational reliability.

MOH operates a diverse fleet of CNC machines and already collects selected production information, but the current workflow requires manual interpretation of dashboards and scattered data sources. As a result, decision-making on throughput losses, idle time, and energy waste depends heavily on expert availability and on time-consuming data analysis. This creates a practical gap between data visibility and operational action, especially when rapid responses are required for scheduling changes, tooling issues or unexpected downtime.

A second challenge is sustainability-driven competitiveness. Automotive customers increasingly require evidence of energy and resource efficiency, yet energy consumption is often monitored at a high level rather than per machine or per part. Without fine-grained, explainable insights, it is difficult to identify specific energy loss mechanisms such as long idle periods with high standby power, suboptimal sequencing of jobs, or recurring micro-stoppages that increase energy per part.

Finally, the adoption of AI-based assistance in manufacturing depends on trust, usability and compliance. Recommendations must be transparent and kept under human oversight, and the technical solution must integrate with existing shop-floor infrastructure using open interfaces. The consortium must therefore deliver a robust, interoperable system and a conversational assistant that production staff can rely on in daily work without creating new operational burdens.

Objective 1:

Deploy a unified, real-time data acquisition pipeline that connects MOH’s ten CNC machines to a central platform through industrial edge controllers and energy-monitoring sensors, enabling reliable capture of machine state, downtime events, cycle counts and power consumption using open protocols and formats.

Objective 2:

Develop and validate an analytics layer that computes at least four core KPIs, including OEE, machine utilization, idle time ratio and energy consumption per part, and that can quantify measurable improvements in productivity and energy intensity during the experiment.

Objective 3:

Implement a WASABI-aligned Digital Intelligent Assistant that automatically generates daily natural-language reports and supports interactive queries for production personnel, and package the resulting assistant skill for distribution through a WASABI White Label Shop instance to enable replication by other manufacturing SMEs.

The experiment is carried out in the metal machining sector, with MOH d.o.o. producing precision-turned and milled parts for automotive customers. The pilot will take place at MOH’s production site in Skofja Loka, Slovenia, in a real operational environment with live production constraints, including machine availability, shift schedules, safety requirements and IT/OT security policies.

Target users are production managers, shift leaders and machine operators who need rapid, interpretable insights into machine efficiency and energy performance. Key stakeholders include MOH management responsible for productivity and sustainability, and Inkolteh as the technology provider aiming to productize the solution as an advanced ccLEAP module. DIH Slovenia supports alignment with trustworthy AI and regulatory expectations, and will use the results as a demonstrator for SME digital transformation and compliance best practices.

Integration constraints include the need to interface with heterogeneous CNC equipment from multiple vendors and to operate reliably in an industrial network. To address this, the architecture relies on standard industrial interfaces (such as Modbus and OPC UA) and IoT messaging (MQTT) with open data representations (e.g., JSON payloads and REST APIs). The assistant will be deployed with containerized components (Docker-based stack) to support maintainability and reproducibility, and all user-facing outputs will remain under human oversight, with clear traceability from recommendations to underlying KPI evidence.