The progressive digital transformation of the industrial sector, guided by the principles of Industry 5.0, promotes a human-centric vision in which Artificial Intelligence technologies support operators by simplifying complex tasks and improving efficiency. Within this context, after-sales technical support represents a strategic area where intelligent systems can reduce downtime and enhance service quality. This thesis presents the design and implementation of a prototype of a Retrieval-Augmented Generation (RAG) chatbot developed within the COMMIS 5.0 project (Chatbot e interfacce umano-centriche per l’OttiMizzazione della Manifattura e l’Innovazione nei Servizi 5.0), funded by the Veneto Region and promoted by the Improvenet network. The industrial partner IMESA S.p.A., a manufacturer of professional laundry machines, provided the use case and technical documentation. The prototype was developed in collaboration with Statwolf S.r.l., whose technical guidance and expertise in data-driven solutions were essential to the system’s design and implementation. The resulting chatbot is intended to assist technicians and customers by offering context-aware answers related to machine operation and troubleshooting. A preprocessing pipeline was implemented to extract, clean, and structure the textual content from multilingual PDFs of IMESA’s manuals, producing Markdown files. This enabled the construction of a vector database for semantic retrieval, ensuring precise document alignment during query processing. The RAG pipeline, encapsulated in Python, combines document retrieval and LLM-based answer generation, while a Flask web interface enables intuitive interaction. The interface features a chat window, contextual metadata, multilingual support, feedback collection, and error handling, providing a practical and user-friendly experience for technicians. The developed prototype demonstrates the feasibility of conversational AI for industrial after-sales services. Future developments will include expanding the knowledge base with additional data sources, integrating visual materials, and deploying the chatbot as a fully operational web application accessible beyond the local environment.
La progressiva trasformazione digitale del settore industriale, fondata sui principi dell’Industria 5.0, promuove una visione human-centric in cui le tecnologie di Intelligenza Artificiale supportano gli operatori semplificando compiti complessi e migliorando l’efficienza. In questo contesto, l’assistenza tecnica post-vendita rappresenta un ambito strategico in cui i sistemi intelligenti possono ridurre i tempi di fermo macchina e migliorare la qualità del servizio. La presente tesi descrive la progettazione e l’implementazione di un prototipo di chatbot basato su Retrieval-Augmented Generation (RAG), sviluppato all’interno del progetto COMMIS 5.0 (Chatbot e interfacce umano-centriche per l’OttiMizzazione della Manifattura e l’Innovazione nei Servizi 5.0), finanziato dalla Regione Veneto e promosso dalla rete Improvenet. Il partner industriale IMESA S.p.A., produttore di macchine professionali per lavanderie, ha fornito lo use-case e la documentazione tecnica. Il prototipo è stato sviluppato in collaborazione con Statwolf S.r.l., la cui guida tecnica e competenza in soluzioni data-driven sono state fondamentali per la progettazione e l’implementazione del sistema. Il chatbot risultante è pensato per assistere tecnici e clienti offrendo risposte contestuali relative al funzionamento delle macchine e alla risoluzione dei guasti. È stata implementata una pipeline di preprocessing per estrarre, pulire e strutturare il contenuto testuale proveniente dai manuali multilingue di IMESA in formato PDF, generando file Markdown. Questo ha permesso la costruzione di un database vettoriale per il recupero semantico, garantendo un allineamento preciso dei documenti durante l’elaborazione delle query. La pipeline RAG, realizzata in Python, combina il recupero dei documenti con la generazione di risposte basata su modelli linguistici di grandi dimensioni (LLM), mentre un’interfaccia web sviluppata con Flask consente un’interazione intuitiva. L’interfaccia include una finestra di chat, metadati contestuali, supporto multilingue, raccolta di feedback e gestione degli errori, offrendo un’esperienza pratica e user-friendly per i tecnici. Il prototipo sviluppato dimostra la fattibilità dell’applicazione dell’Intelligenza Artificiale conversazionale nei servizi post-vendita industriali. Gli sviluppi futuri prevedono l’espansione della base di conoscenza con ulteriori fonti di dati, l’integrazione di materiali visivi e la distribuzione del chatbot come applicazione web completamente operativa, accessibile oltre l’ambiente locale.
Chatbot Multimodale Basato su RAG per Servizi di Supporto Avanzati
RIGONI, MARIA CHIARA
2024/2025
Abstract
The progressive digital transformation of the industrial sector, guided by the principles of Industry 5.0, promotes a human-centric vision in which Artificial Intelligence technologies support operators by simplifying complex tasks and improving efficiency. Within this context, after-sales technical support represents a strategic area where intelligent systems can reduce downtime and enhance service quality. This thesis presents the design and implementation of a prototype of a Retrieval-Augmented Generation (RAG) chatbot developed within the COMMIS 5.0 project (Chatbot e interfacce umano-centriche per l’OttiMizzazione della Manifattura e l’Innovazione nei Servizi 5.0), funded by the Veneto Region and promoted by the Improvenet network. The industrial partner IMESA S.p.A., a manufacturer of professional laundry machines, provided the use case and technical documentation. The prototype was developed in collaboration with Statwolf S.r.l., whose technical guidance and expertise in data-driven solutions were essential to the system’s design and implementation. The resulting chatbot is intended to assist technicians and customers by offering context-aware answers related to machine operation and troubleshooting. A preprocessing pipeline was implemented to extract, clean, and structure the textual content from multilingual PDFs of IMESA’s manuals, producing Markdown files. This enabled the construction of a vector database for semantic retrieval, ensuring precise document alignment during query processing. The RAG pipeline, encapsulated in Python, combines document retrieval and LLM-based answer generation, while a Flask web interface enables intuitive interaction. The interface features a chat window, contextual metadata, multilingual support, feedback collection, and error handling, providing a practical and user-friendly experience for technicians. The developed prototype demonstrates the feasibility of conversational AI for industrial after-sales services. Future developments will include expanding the knowledge base with additional data sources, integrating visual materials, and deploying the chatbot as a fully operational web application accessible beyond the local environment.| File | Dimensione | Formato | |
|---|---|---|---|
|
Rigoni_Maria_Chiara.pdf
Accesso riservato
Dimensione
2.49 MB
Formato
Adobe PDF
|
2.49 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/96066