EXPERIMENT OVERVIEW
AVAROS (AI-Voice-Assistant-Driven Resource-Optimized Sustainable Manufacturing) will design, implement, and validate an OVOS-based digital assistant to support operational decision-making for improved resource efficiency in manufacturing and related supply-chain activities. The experiment addresses a common SME problem: decisions that directly influence electricity intensity, material losses, and CO₂-eq performance are still often made with fragmented data, manual spreadsheet work, and tacit know-how. AVAROS will reduce this friction through a human-centered interface that helps users query key performance indicators, understand operational trade-offs, and act on evidence-linked insights.
The solution will be built on RENERYO, ArtiBilim’s supply-chain-oriented optimization backbone, which already consolidates supplier, process, and batch information into actionable KPIs. AVAROS will add a conversational and “what-if” decision layer while keeping RENERYO as the trusted data and KPI source for consistent calculations, traceability, and reporting. Within the WASABI stack, AVAROS will integrate OVOS for voice and text interaction, DocuBoT for document-grounded answers based on procedures, specifications, and pilot documentation, and PREVENTION for anomaly and risk detection in operations and supply flows. The full setup will be packaged with Docker-Compose so that deployment is portable, repeatable, and easier for other SMEs to adopt.
AVAROS will be developed for practical daily use by engineers and planners. Typical use cases will include supplier and material planning, including tracking specifications, lead times, defect trends, and substitution options; production scheduling and changeovers, including anticipating electricity-intensity spikes and reducing peak-tariff exposure; and inventory and scrap control, including identifying at-risk materials earlier and linking lot quality to rework or scrap. Users will interact with the assistant through natural-language voice or text queries and will receive grounded answers, linked evidence, and decision-support suggestions instead of static dashboard views alone.
Implementation will follow a staged validation pathway. ArtiBilim’s own plastics/toy operations will serve as the needs-analysis and development testbed in the first phase. This will be followed by deployment at an EDIH MIDAS-designated pilot factory and validation activities supported through the Digital Transformation Centre (DDM). By the end of the experiment, AVAROS targets at least an 8% reduction in electricity per unit, at least a 5% improvement in material efficiency across the supply chain, and at least a 10% decrease in CO₂-eq versus baseline. To support replication beyond the pilots, the solution will be delivered as a shop-ready dockerized package and published on the WASABI White-Label Shop together with anonymized implementation learnings that can shorten time-to-value for future SME adopters.
The experiment addresses a set of tightly connected operational and technical challenges that are common in energy- and supply-chain-intensive manufacturing. In these environments, sustainability performance depends on many small daily decisions, yet those decisions are often slow, fragmented, and heavily dependent on individual staff knowledge. Factories need to react to changing raw material specifications, supplier lead times, machine mixes, and tariff conditions while also trying to control electricity intensity, material losses, scrap, rework, and CO₂-eq reporting. When these factors are managed through spreadsheets and disconnected data sources, response times become slower and decision quality becomes less consistent.
Before implementation, the main problem is not the absence of data but the way it is organized and accessed. Relevant signals are dispersed across ERP and MES exports, shop-floor measurements, supplier declarations, and supporting documents. This creates a situation in which engineers and planners must manually consolidate information before acting. The result is delayed decision cycles, lower repeatability across shifts or teams, weaker traceability, and higher cognitive load on staff. In volatile conditions, this also leads to avoidable energy cost per unit, inefficient production operations, longer reporting cycles, and missed opportunities for earlier anomaly detection.
A second challenge is transferability. Many digital support tools remain site-specific or prototype-level and are difficult for other SMEs to reuse. AVAROS therefore addresses not only operational improvement at pilot level, but also the challenge of packaging a reusable assistant that can be deployed with limited adaptation, open interfaces, and a clear installation pathway. This is especially important in the WASABI context, where the value of the experiment depends on moving beyond a one-off demonstrator toward a portable, shop-ready solution that other SMEs can realistically test and adopt.
Objective 1:
To design and implement an OVOS-based, human-centered digital assistant that supports operational decision-making for improved resource efficiency in manufacturing and related supply-chain activities. The assistant will help engineers and planners access KPIs, receive grounded explanations, and explore “what-if” scenarios through natural-language interaction, reducing dependence on fragmented spreadsheets and manual interpretation.
Objective 2:
To integrate OVOS, DocuBoT, PREVENTION, and RENERYO into a portable, dockerized architecture that combines conversational interaction, document-grounded support, and anomaly detection in one repeatable deployment package. This objective focuses on building a technically robust and reusable solution with open interfaces, traceable data flows, and compatibility with real industrial environments.
Objective 3:
To validate the solution in ArtiBilim’s own operations and in an EDIH MIDAS-supported pilot environment, demonstrating measurable improvements in electricity intensity, material efficiency, and CO₂-eq performance while preparing the solution for replication through publication on the WASABI White-Label Shop. This objective ensures that AVAROS produces both pilot-level evidence and a clear pathway for wider SME uptake.
AVAROS is positioned in energy- and supply-chain-intensive discrete manufacturing, starting from ArtiBilim’s plastics/toy production context and designed to transfer to automotive and textile supplier environments. The experiment therefore sits in a real industrial setting rather than a purely laboratory environment. Its primary use case concerns the daily interaction between operational staff and resource-efficiency data, especially where production, supplier performance, quality outcomes, and energy use need to be interpreted together rather than in isolation.
The experiment will be carried out in Türkiye. The first implementation setting will be ArtiBilim’s own manufacturing operations, which will serve as the needs-analysis and development validation testbed. This will allow the team to define use cases, assess data readiness, validate connectors, and stabilise the assistant in a live operational environment. In the second phase, the solution will be deployed in parallel through an EDIH MIDAS-designated pilot factory and supported activities at DDM, where transferability, onboarding, and validation workshops can be carried out with external stakeholders under realistic industrial routines.
The main target users are engineers, planners, and operational decision-makers who need faster access to trusted KPIs and grounded recommendations. Other relevant stakeholders include pilot-factory operators, management teams interested in productivity and sustainability improvements, EDIH MIDAS as the adoption and replication partner, and future SME adopters accessing the solution through the WASABI Shop. The experiment also assumes integration with existing data sources such as ERP, MES, IIoT, supplier files, and documentation repositories. Key constraints include data availability and quality, connector readiness, pilot access coordination, and the need to maintain GDPR-aware governance and AI-risk mitigation measures throughout implementation. The proposal also frames the solution as limited-risk industrial decision support with human oversight maintained throughout.
EXPECTED IMPACT
EXPECTED IMPACT
The expected impact of AVAROS is technical, operational, economic, and replication-oriented at the same time. At technical level, the experiment is expected to show that conversational access to resource-efficiency signals can shorten the loop from detection to action in real manufacturing operations. Instead of forcing users to interpret separate dashboards, spreadsheets, and documents, the assistant will provide a single interaction layer that helps planners and engineers move more quickly from data to decision. Expected operational effects include reduced reporting latency, earlier visibility of supplier and process deviations, more consistent “what-if” exploration, and lower cognitive load on users.
At performance level, the experiment will be considered successful if it demonstrates measurable improvement against defined baselines. The main targets are at least an 8% reduction in electricity intensity per unit output, at least a 5% improvement in material efficiency across the supply chain, and at least a 10% decrease in CO₂-eq intensity versus baseline. Baselines will be frozen at the beginning of deployment using available operational and supply-chain records, including meter readings and/or utility invoices mapped to production output, alongside documented emission factors for traceable CO₂-eq calculations.
At business level, AVAROS is expected to strengthen RENERYO by adding a native digital intelligent assistance layer based on OVOS, DocuBoT, and PREVENTION. This should improve usability, accelerate onboarding, increase day-to-day use, and create clearer links between actions and outcomes for users. For ArtiBilim as a manufacturing SME, the experiment also has direct internal value through optimization of supplier selection, better handling of peak-tariff exposure, and earlier anomaly detection that can reduce scrap and rework. The proposal also notes that a mid-size plant adopting the solution could achieve approximately %16 in annual energy savings, although that figure should be used carefully as an indicative value rather than a guaranteed result.
At replication level, success also depends on whether the solution becomes reusable beyond the pilots. AVAROS is therefore expected to produce a dockerized release, installation checklist, sample configuration, a minimal getting-started dataset, and anonymized resource-efficiency learnings for publication on the WASABI White-Label Shop. Dissemination and exploitation targets include engagement of SMEs through EDIH MIDAS channels, at least five SME onboarding calls via EDIH, and at least two external proof-of-concept opportunities within three months of listing. These outputs will show that the experiment has moved beyond a one-site demonstrator toward a low-friction replication pathway for other SMEs.