Artificial Intelligence, Multi-Agent Systems
Work description
The scholarship's area of training is Multi-Agent Systems and Human-Centred Artificial Intelligence (Human-Centred AI), focusing on the study, research and development of agents for the construction, customisation and continuous adaptation of dashboards and KPIs in an organisational context. The goal is to advance the state of the art in architectures and methods that enable the automatic generation of high-value dashboards for the business, linking data to objectives, context and decisions, and supporting the exploration, explanation and validation of KPIs by different user profiles (managers, analysts, operational teams). In this context, socio-technical and human-centric dimensions are essential: the aim is for agents to operate in a transparent, controllable manner, aligned with real practices of governance, data literacy and organisational responsibility. In particular, the scholarship aims to investigate a family of specialised agents (e.g., goal discovery agents, semantic modelling of KPIs, visualisation recommendation, explanation and narrative, persona-based personalisation, and feedback-based adaptation) that integrate into an existing multi-agent architecture, extending it with advanced capabilities for: (i) business connection (strategy, processes, objectives, KPI contracts, ownership); (ii) automatic dashboard composition (selection of metrics, layout, visualisations, alerts, hierarchies, and drill-down); (iii) dynamic and contextual adaptation (changing priorities, seasonality, events, data and user changes); and (iv) explainability, trust and human validation (justification, traceability and control mechanisms). In addition, the aim is to explore robust mechanisms for integration and governance (catalogue, lineage, quality, access, auditing), ensuring that automatically generated dashboards are accurate, relevant, auditable and actionable. Specifically, the main activities to be carried out by the scholarship holder are: • Analyse the state of the art in agents and multi-agent architectures applied to analytics and BI (including LLM-agents, planning and tool-use, semantic RAG, visualisation recommendation systems, conversational agents for BI, and human-in-the-loop approaches), focusing on current limitations in the generation of value-oriented dashboards and the explanation of KPIs; • Identify and specify advanced requirements (functional and non-functional) for a dashboard/KPI-oriented agent layer: alignment with business objectives, personalisation by persona, compliance constraints, governance and auditing, robustness to imperfect data, and UX/explainability requirements; • Propose formal/semantic models and representations to link business-data-KPI (e.g., KPI contracts, metric ontologies/taxonomies, goal-metric-action models, organisational context, and user profiles), enabling agents to reason about relevance, impact, and trade-offs; • Develop new agents that implement functionalities such as automatic dashboard construction (specification generation, visualisation selection, layout and analytical storytelling), question-driven exploration (‘why did it go up/down?’, ‘what explains the variation?’, ‘where to act?’), explanation and narrative of KPIs (justifications, breakdowns, drivers, comparisons, uncertainty), personalisation and adaptation (preferences, tasks, literacy, device, temporal context, behaviour, and feedback); • Investigate coordination and security mechanisms between agents and tools (orchestration, goal negotiation, memory and context management, access policies), including strategies to prevent hallucinations and metric errors, automatic validation (statistical/semantic checks), and trust calibration; • Integrate agents into an existing architecture, defining interfaces, interaction patterns, connectors to organisational sources (catalogue • Validate in realistic scenarios, through industrial use cases (e.g., commercial performance, operation, finance, supply chain), conducting comparative experiments with traditional approaches and evaluating performance, robustness, generalisation, and organisational impact; • Collaborate in technical-scientific production, including articles, reports, architecture and agent documentation, experimental reproducibility, and writing activity reports, systematising results and recommendations for industrial adoption and governance of agent-based solutions.
Academic Qualifications
- Master's degree in computer engineering, information systems, or related field;
Minimum profile required
- Average grade in bachelor's and master's degrees higher than 14.
Preference factors
- Fluency in Portuguese. - Experience in Business Intelligence/Analytics - Experience with agent frameworks
Application Period
Since 05 Mar 2026 to 18 Mar 2026
Centre
Industrial & Systems Engineering and Management