LiquioAI – Trusted AI Assistant for Public Services
LiquioAI is a deep-tech conversational assistant for public services, built on large language models but constrained by strict governance, evaluation and depersonalisation. Integrated with the Liquio low-code platform via standard interfaces and MCP tools, it guides citizens through complex procedures (e.g. benefits, certificates, applications) and produces audit-ready outputs that public administrations can trust. LiquioAI is designed as a civil-only AI assistant for public services in the EU and Ukraine. It runs on EU/Associated cloud or on-premise infrastructure, enforces depersonalisation before any LLM call and restores identity only at the edge. An Evaluation Harness with gold and adversarial scenarios measures task success, time to first visible answer, grounding, hallucination and safety. Accessibility (WCAG 2.2 AA / EN 301 549) and security (threat modelling, RBAC, logging, penetration testing) are built in from the start, aligned with GDPR and the EU AI Act requirements for high-risk systems. The project will advance LiquioAI from TRL 4 to TRL 6–7 by designing and validating the architecture, running sandbox and operational pilots with BELDOC (Belgium) and/or any of the two Ukrainian public administrations, and/or one additional EU/Associated administration, and delivering an evidence-based governance pack (AI Risk File, DPIA, evaluation and security reports). Independent external validation (e.g. by a university or test lab) will support credibility. The result is a commercially deployable, trustworthy AI assistant for citizens that preserves control, accountability and compliance for public authorities in the EU and Ukraine.
4/6 complete
Imported from OpenAIRE Ukraine | Grant 101313057.
Open sourceOverview
Grant agreement 101313057 | LiquioAI
Open sourceLiquioAI is a deep-tech conversational assistant for public services, built on large language models but constrained by strict governance, evaluation and depersonalisation. Integrated with the Liquio low-code platform via standard interfaces and MCP tools, it guides citizens through complex procedures (e.g. benefits, certificates, applications) and produces audit-ready outputs that public administrations can trust. LiquioAI is designed as a civil-only AI assistant for public services in the EU and Ukraine. It runs on EU/Associated cloud or on-premise infrastructure, enforces depersonalisation before any LLM call and restores identity only at the edge. An Evaluation Harness with gold and adversarial scenarios measures task success, time to first visible answer, grounding, hallucination and safety. Accessibility (WCAG 2.2 AA / EN 301 549) and security (threat modelling, RBAC, logging, penetration testing) are built in from the start, aligned with GDPR and the EU AI Act requirements for high-risk systems. The project will advance LiquioAI from TRL 4 to TRL 6–7 by designing and validating the architecture, running sandbox and operational pilots with BELDOC (Belgium) and/or any of the two Ukrainian public administrations, and/or one additional EU/Associated administration, and delivering an evidence-based governance pack (AI Risk File, DPIA, evaluation and security reports). Independent external validation (e.g. by a university or test lab) will support credibility. The result is a commercially deployable, trustworthy AI assistant for citizens that preserves control, accountability and compliance for public authorities in the EU and Ukraine.