Federated clinical AI network
Diagnostic or prognostic models trained across multiple hospitals without any patient record leaving its institution, federated learning hardened with differential privacy and, where verifiability is required, ZKML proof of correct inference.
- Our role: privacy architecture, FL infrastructure, model engineering
- KPI framework: model quality vs. centralised baseline, privacy budget (ε), cross-site generalisation
- TRL path: 4 → 7 across three annual pilot rounds
Edge vision quality inspection
In-line defect detection on production lines using edge-deployed vision models - trained with limited labelled data, optimised with TensorRT, and integrated with the factory MES so operators act on findings in real time.
- Our role: model development, edge deployment, MES integration
- KPI framework: detection precision/recall, false-reject rate, inference latency, scrap reduction
- TRL path: 5 → 8 with two industrial end-users
Autonomous inspection robotics
Ground robots with arm-mounted sensors for inspecting energy and transport infrastructure: autonomous mission planning, on-board defect perception and automated reporting into the asset-management system.
- Our role: ROS 2 autonomy stack, on-board perception, mission orchestration
- KPI framework: inspection coverage, defect-detection accuracy, inspection cost/time vs. manual baseline
- TRL path: 4 → 7, simulation-first then site trials
Wildfire & environmental early warning
Multi-source early-warning demonstrator combining camera towers, aerial patrols and satellite feeds with edge AI detection, designed for Mediterranean wildfire conditions we know first-hand from Greece.
- Our role: detection models, edge deployment, alert-fusion platform
- KPI framework: time-to-detection, false-alarm rate, area under surveillance per euro
- TRL path: 5 → 8 with civil-protection end-users
Sovereign RAG for public administration
A retrieval-augmented assistant over legislation, case files and administrative procedures, open-weight LLMs fine-tuned for the national language, deployed on-premises, with citation grounding and full audit logging.
- Our role: corpus pipeline, fine-tuning, RAG platform, evaluation
- KPI framework: answer groundedness, task completion time, staff satisfaction, zero data egress
- TRL path: 5 → 8 within a single programme cycle
Digital twin for energy assets
A live digital twin of buildings, district networks or renewable assets that fuses IoT telemetry with predictive models, forecasting demand and failures, and closing the loop with optimisation recommendations.
- Our role: twin platform, predictive analytics, decision-support dashboards
- KPI framework: forecast accuracy, energy savings (%), unplanned-downtime reduction
- TRL path: 4 → 7 with utility or ESCO partners