At DataWeek 2026 in Oslo, AIXPERT, represented by Technical Coordinator NOVELCORE, presented a clear and compelling vision: trustworthy AI must be explainable, accountable, and transparent by design. As part of a joint session on Human Oversight and Hybrid Intelligence – bringing together EU-funded projects including Themis 5.0, FAITH, and AI Mimer – the project showcased an agentic, multi‑layer GenAI backbone built specifically to meet these expectations in regulated, high‑impact domains such as healthcare, recruitment, manufacturing, educational robotics, and creative industries.
Addressing the Core Problem: The Black‑Box Barrier
AIXPERT’s starting point is the persistent issue of black‑box AI. As models grow in capability, their internal reasoning becomes harder for humans to understand or audit. The project argues that high‑risk applications cannot rely on opaque systems, regardless of performance. Instead, they require architectures that make reasoning observable, evidence‑grounded, and open to human intervention.
A Multi‑Layer Solution for Trustworthy AI
The AIXPERT solution consists of several layers:
Cognitive Foundation Layer
AIXPERT begins with models that are not just powerful, but interpretable and aligned with their deployment context. This includes:
- explainable multimodal models
- domain fine‑tuning
- bias mitigation and cultural sensitivity
- RLHF‑based alignment
This layer provides the technical substrate for responsible AI reasoning.
Dialogue Mediation Layer
The central innovation is AIXPERT’s supervisory mediation system that transforms a raw LLM into a traceable reasoning engine. It includes:
- an orchestrator for task decomposition
- knowledge‑graph grounding
- RAG‑based evidence retrieval
- traceable reasoning chains
- trust‑decay scoring to flag low‑confidence outputs
- LLM‑as‑judge evaluation for internal quality control
This layer is where transparency is operationalised.
Agent‑World Interface Layer
For real‑world deployment, the system integrates with external environments through:
- controlled tool execution
- links to EHRs or other APIs
- an immutable audit trail
- policy‑as‑code rules
- mandatory guardrails and action approvals
Every decision becomes verifiable and traceable.
Human Oversight Checkpoints
AIXPERT emphasises that explainability is meaningless without human oversight. Its framework formalises human checkpoints:
- collaborative planning
- pre‑action approval
- post‑execution review
- clinical or expert confirmation
- escalation triggers for ambiguous or risky cases
Oversight remains present, structured, and empowered.
Outcomes: A System Built for Trust
AIXPERT’s contribution to Oslo Data Week highlights that trustworthy AI is not a single feature; it is a system outcome that emerges from architecture, governance, and human collaboration. The project delivers:
- Explained and auditable decisions powered by evidence and reasoning trails
- Human‑controlled actions, enabling override, approval, or delegation
- A design philosophy centered on Explainability, Accountability, and Transparency
In a forum focused on demystifying AI and moving “beyond the black box,” AIXPERT offered a concrete, multilayer blueprint for how next‑generation AI systems can earn trust—not through promises, but through design.