Te Whatu Ora - Nelson Hospital
Health Care
AI Agent
Digital Diligence Limited (DDL) developed a Retrieval-Augmented Generation (RAG) AI agent proof of concept (PoC) for Te Whatu Ora - Nelson Hospital. This AI agent leverages OpenAI's LLM API and connects to curated clinical data, assisting non-specialist frontline staff, such as general practitioners, in obtaining insights and potential treatment pathways based on specific symptoms.
01
OVERVIEW
Digital Diligence Limited (DDL) developed a Retrieval-Augmented Generation (RAG) AI agent proof of concept (PoC) for Te Whatu Ora - Nelson Hospital. This AI agent leverages OpenAI's LLM API and connects to curated clinical data, assisting non-specialist frontline staff, such as general practitioners, in obtaining insights and potential treatment pathways based on specific symptoms. Due to privacy and confidentiality agreements, specific elements of this project cannot be disclosed. The PoC serves as a technological feasibility study rather than a fully deployed clinical tool.
02
THE challenge
Frontline healthcare providers, particularly general practitioners, frequently encounter cases that require specialist input. However, access to specialists may be delayed due to demand, resource constraints, or geographical limitations. The objective of this proof of concept was to explore how AI-driven assistance could support decision-making by providing rapid, relevant, and evidence-based insights without replacing clinical judgment.
03
Solution Delivered by Digital Diligence Limited
DDL designed a RAG-based AI agent, which has been trialled using combinations of Graphlit, Notebook LLM, OpenAI and Python code with a database of curated clinical guidelines. The AI agent functions as a knowledge retrieval system, ensuring responses are:
-
Clinically Relevant – Derived from validated medical sources.
-
Contextually Aware – Providing guidance based on symptomatology rather than generic responses.
-
Interactive & Adaptive – Designed to refine responses based on follow-up queries.
The AI front-end was tailored to fit the workflow of general practitioners, allowing them to input patient symptoms and receive structured insights into potential diagnostic pathways and treatment considerations. This approach ensures clinicians remain in control of decision-making while benefiting from AI-augmented insights.
03
implementation process
The PoC included:
-
Data Curation & Structuring: DDL worked with subject matter experts to structure a dataset of clinical guidelines specific to the target condition(s) which comprised hundred of pages of text to be chunked into managable sections of knowledge.
-
RAG Architecture Development: Evaluating and implementing different LLMs and API to support best-fit, with a retrieval system to fetch and synthesise relevant data in real time.
-
User Interface Design: A simple, intuitive web-based front end allowing practitioners to query the AI agent effortlessly.
Testing & Validation: Running test cases to ensure the AI's responses aligned with expected clinical knowledge, with feedback loops for refinement.
04
Key Outcomes
-
Feasibility Evaluation: The AI agent can now be tested in limited clinical contexts, retrieved relevant clinical information, demonstrating the viability of RAG-based solutions in medical settings.
-
Improved Accessibility: General practitioners could rapidly access structured insights on complex conditions without requiring direct specialist consultation.
-
Enhanced Decision Support: While not a replacement for human expertise, the tool is designed to provide valuable supplementary insights for consideration in clinical decision-making.
05
Conclusion
Digital Diligence Limited successfully demonstrated the potential of RAG AI technology in a clinical decision-support context. While this proof of concept is not yet in active clinical use, it lays the groundwork for future AI-driven tools that could enhance frontline healthcare delivery. The project highlights the role AI can play in augmenting medical expertise, ultimately improving patient outcomes through intelligent, data-driven insights.