AIM-SAFE

Safe, explainable multi-agent AI for clinical decision support

Consortium26 partners · 14 countries

RoleTechnical Partner — Co-Pilot Interface & XAI

ProgrammeHORIZON EUROPE

Grant reference101287214

Our role

Technical Partner — Co-Pilot Interface & XAI

We build the clinician-facing Co-Pilot interface and the XAI and trust-calibration dashboards that make the system's recommendations understandable and auditable. We also lead the multi-site data inventory, gap analysis and governance, bringing our NLP, knowledge-management and agentic RAG work from AIXPERT.

Impact and outcomes

Targeting validation in at least 4 hospitals across at least 3 EU countries, in geriatrics and oncology
Targeting at least 80% sensitivity and specificity for polypharmacy adverse-drug-event prediction on historical data
Targeting at least 90% of unsafe AI outputs intercepted during hospital pilots
Targeting explanations in at least 4 EU languages and a 20% or greater gain in clinician trust

The challenge

Medication errors affect more than 10% of hospitalised patients, and adverse drug events cause 200,000 deaths a year in Europe. Clinicians must weigh drug interactions, pharmacogenomics and changing guidelines for multimorbid patients within short consultations. Current decision-support systems are siloed and rule-based, and they flood clinicians with undifferentiated alerts.

Our approach

AIM-SAFE's Safety Co-Pilot orchestrates small language model agents that run locally in hospitals and improve through federated learning, without centralising patient data. A Safety Governance Layer screens every recommendation for hallucinations, dosing errors and non-compliance before a clinician sees it. We build the interface and dashboards through which clinicians query the system, see traceable rationales and calibrate their trust.