BIRMINGHAM, England: We are excited to be extending our work with NVIDIA to help advance the safe and trusted deployment of agentic AI.

This builds upon our launch in June this year of Care Navigator, the UK’s first private sovereign healthcare LLM trained on NHS primary care data. Through this work, we demonstrated what can be achieved when NVIDIA’s models and AI infrastructure are combined with OneAdvanced’s deep domain expertise. Using NVIDIA Nemotron open models and NeMo open libraries, OneAdvanced developed a sovereign healthcare model tuned for the specific needs of NHS primary care-achieving significantly higher accuracy than the GP control group in benchmark testing, while delivering up to 150x lower inference costs than leading frontier models.

Care Navigator demonstrates the value of specialised open-weight models for tasks such as NHS patient triage. Tuning models to a specific domain can improve relevance and accuracy while reducing inference costs, creating capacity to invest further in evaluation, security, safety and R&D.

Turning to Agents

Building on this work with a specialised model, we are now exploring how AI can support more autonomous workflows across OneAdvanced IQ. Our platform users want agents that can work across its diverse workflows and data while respecting each user’s permissions. We are using NVIDIA technologies to help make these capabilities safer and more secure.

To support this work, we have adopted NVIDIA NeMo Gym to evaluate the models and agents underpinning OneAdvanced IQ, our Intelligent System of Work platform. Our testing goes beyond the quality of an agent’s response to examine whether it takes the right action, uses the appropriate tools and respects the limits of its authority. We continue to expand upon our systems of AI evaluation so that our platform users will soon be able to safely leverage both the benefits of open-models and agentic AI.

This becomes increasingly important as AI moves towards taking actions on behalf of users. For enterprises-particularly companies like ours that serve highly regulated sectors-unlocking the full potential of agentic AI will depend on being able to demonstrate that agents behave safely and as intended.

Fonte: Business Wire