What is NOA?
NOA is an LLM-based AI care assistant developed at Mentech Innovation. It is designed to support care professionals in their daily work through context-aware, intelligent guidance, helping them navigate complex care situations, documentation tasks, and decision support.
NOA is built entirely on private AI infrastructure, meaning no data leaves the organisation’s environment. This is a deliberate architectural choice driven by the sensitivity of healthcare data and the strict regulatory requirements that apply in the Dutch care sector.
Why private infrastructure?
Healthcare organisations handle highly sensitive personal data: medical records, behavioural observations, and care plans, governed by NEN 7510 (the Dutch standard for information security in healthcare) and the GDPR. Sending this data to commercial LLM APIs is not viable from a compliance or trust perspective.
Mentech’s approach is to run open-weight LLMs on-premises or in a private cloud environment, with full control over data flows, audit logging, and access management. NOA is ISO 42001 certified, meaning its AI processes meet the international standard for AI management systems, with transparency, risk management, and accountability built in.
Technology
- LangChain for building and managing the reasoning pipelines, tool use, and retrieval-augmented generation (RAG) flows
- Cloud hosted with a specific focus on EMEA region data processing and LLM serving, ensuring data stays within European jurisdiction
- Prompts customised based on care-specific terminology, protocols, and the personal interests and context of individual clients
- Connected to the care organisation’s ECD (electronic client dossier) and other care planning tools via secure APIs
- Full audit trail, explainability logging, and role-based access control to meet ISO 42001 and NEN 7510 requirements
- A desktop-based HMI with a personalised avatar, video, audio, and entertainment content tailored to the individual resident
My role
As CTO of Mentech, I initiated and led the NOA project from architecture through deployment. This included selecting the underlying model stack, designing the private infrastructure pattern, and establishing the compliance framework that enabled adoption in regulated care environments.