Contractació dels serveis per al disseny, desenvolupament i validació, en format de prova de concepte, d'un assistent virtual en l'àmbit de la salut i el benestar, orientat a la ciutadania, basat en Intel·ligència Artificial Generativa.

The contract timeline is as follows:

• Contract Start Date: The day after the contract is formalized.
• Contract End Date: December 31, 2026.
• Total Duration: From the day after contract formalization until December 31, 2026. The contract is not subject to extension.
• Key Milestones/Phases:
• Start of the project: Delivery of the work plan including timeline, resource allocation, and risk matrix.
• Before development: Submission of the design report, which includes a comparative analysis of solutions, the approved technical and functional proposal, and the catalog of evaluation indicators.
• Before the execution of simulated interactions: Deployment of the functional system in a controlled validation environment, along with technical documentation of the architecture.
• Before the evaluation team's training: Delivery of the concept proof validation protocol (including the design of simulated interactions, indicators, and data collection and analysis procedures) and the system user manual for the evaluation team with training support materials.
• End of the project: Submission of the concept proof evaluation report (containing quantitative and qualitative results, detected incidents, improvement proposals, and scalability study) and the source code with complete technical documentation in an accessible repository.

• Integrable within the website of the Escola de Salut Catalana
• Interaction via written natural language
• Responses based exclusively on official content validated by the Department of Health
• Responses in clear, inclusive, understandable language adapted to general population
• Available in Catalan
• Available in Spanish
• Voice recognition for queries (optional, but if offered, specific security and data protection requirements apply)
• Voice synthesis for responses (optional, but if offered, specific security and data protection requirements apply)
• Voice processing not to involve conservation of recordings
• Voice processing not to involve identification of users
• No generation of clinical diagnoses
• No personalized therapeutic recommendations
• No interpretation of medical tests
• No substitution of healthcare assistance
• Compliance with the National Security Scheme (ENS) regulated by Royal Decree 311/2022
• Compliance with the European Artificial Intelligence Regulation for limited-risk AI systems
• Clear information that the user interacts with an AI system
• Information that content is generated via AI
• Transparency in the general operation of the system
• Avoid designs that may mislead about the non-human nature of the system
• Incorporation of a visible warning that it is a generative AI system
• Explicit declaration of its educational purpose
• Redirection messages to health services for urgent situations, severe symptoms, or health risks
• Reminder that information does not replace professional attention
• Allow structured loading of official documents from the Department of Health
• Guarantee traceability of sources used in each response
• Incorporate mechanisms to avoid responses outside the authorized documentary scope
• Allow immediate deactivation or modification of content if errors are detected
• Provide technical documentation including data processing description
• Provide technical documentation including selection of language models employed
• Provide technical documentation including evaluation techniques used
• Provide transparency and communication protocols for the AI tool's operation
• Informative texts visible in the user interface about the capabilities, limitations, benefits, and risks of the solution
• Comply with WCAG 2.1 AA or higher criteria
• Fully functional on mobile devices
• Compatible with the Canal Salut graphic identity
• Not require prior registration or authentication
• Generative Artificial Intelligence
• Validation protocol for the proof of concept, based on simulated interactions
• Evaluation of the Generative AI tool, including indicators and metrics of reliability, security, usability, acceptance, and satisfaction
• Thematic scope restricted to Healthy eating
• Thematic scope restricted to Healthy physical activity
• Thematic scope restricted to Living smoke-free - tobacco
• Thematic scope restricted to Responsible alcohol consumption
• System must only respond to queries related to the four specified thematic areas
• System must discard queries outside the thematic perimeter and inform the user
• Responses based exclusively on official documentary content from Suma Salut, Canal Salut - Vida Saludable, and Escola de Salut Catalana
• No use of data derived from interactions for system training or re-training
• Deliverables in open formats
• Deliverables in an accessible repository
• Deliverables documented and structured for reuse, maintenance, and evolution by an independent technical team
• Documentation of use, configuration, and dependencies for third-party components, tools, or models (e.g., commercial language models, libraries, external services) where assignment rights are not held by the contractor
• Full delivery of source code
• Full delivery of technical documentation