DEXTRO-LH – Explainable and Adaptive AI Recommendation Framework for Hospital Logistics

DEXTRO-LH is an AI-based recommendation framework designed to optimise hospital logistics, with a particular focus on inventory management, material traceability and operational decision support. The solution combines personalised recommendation models, natural-language interaction, expert agents, knowledge graphs and multimodal information from hospital databases, RFID systems and computer vision.

The framework is being designed as a modular and adaptable platform that can be configured according to different hospital environments and professional profiles, including logistics staff, operating-room personnel, nursing staff and inventory managers. Its recommendation engine combines content-based and collaborative approaches with neural recommendation models and graph-based representations to generate context-aware suggestions.

Modular, service-oriented architecture for hospital logistics recommendation and decision support. Natural-language interaction through the QnA Front, with REST-based communication between frontend and backend services. Dynamic user profiling based on role, organisational position, interaction patterns and contextual information. Recommendation engine combining content-based, collaborative and hybrid approaches, with support for neural and graph-based recommendation techniques. Integration of expert agents, knowledge graphs and contextual reasoning to enrich recommendations. Explainability mechanisms based on XAI, including SHAP, LIME, natural-language explanations and dynamic visualisations. Support for metacognitive reasoning approaches such as LogiCoT and Graph of Thoughts (GoT). Multimodal processing capabilities combining structured hospital data, text, images, RFID and IoT signals. Security and scope validation through dedicated Security Analyzer and Scope Analyzer services. Federated-learning compatibility to enable distributed model training without centralising sensitive hospital data. API-based interoperability with hospital databases, ERP systems, inventory applications, RFID infrastructures and external services. Container-ready deployment using technologies such as Docker, with support for scalable cloud/on-premise environments.

Integration requires access to hospital information systems through documented APIs, database connectors or compatible data-export mechanisms. External systems must provide sufficiently structured and semantically consistent information on inventory, materials, users, roles and operational context. REST/JSON interfaces are the preferred integration mechanism; adapters may be required for legacy systems using proprietary protocols or data formats. Authentication, authorisation and access-control mechanisms must be aligned with the security policies of each hospital. Integration of RFID, IoT or computer-vision components depends on the availability and compatibility of local sensing infrastructure. Local data schemas may require mapping, normalisation or transformation before being consumed by DEXTRO-LH. Federated deployments require compatible local training environments and agreed model-exchange mechanisms between participating nodes. Hospital-specific rules, terminology, user roles and operational processes must be configured before recommendations can be reliably contextualised. Deployment must comply with applicable privacy, cybersecurity and data-governance requirements, particularly when clinical or sensitive operational data are involved. Performance and scalability depend on the available computing infrastructure, especially when multimodal models or local AI inference are enabled.

Hospitals, hospital groups, healthcare logistics departments, operating-room management units, procurement departments, medical-device logistics providers and healthcare technology integrators.

DEXTRO-LH is intended to be reusable as a configurable software framework across different hospitals and healthcare logistics environments. Reuse requires adaptation of connectors, data mappings, user profiles, access rules and domain-specific business logic to the target organisation.

The reusable assets include the modular recommendation framework, API specifications, profiling services, Security and Scope Analyzers, Graph Generator, expert-agent interfaces and explainability components. Hospital-specific datasets, credentials, proprietary integrations and operational rules remain under the control of the corresponding organisation.

Reuse may be provided under a commercial licence or project-specific exploitation agreement defined by DEXTROMEDICA. Third-party libraries, models and services integrated into the platform remain subject to their respective licences and terms of use. Deployments processing personal or sensitive information must comply with GDPR and applicable institutional security and governance policies.