Practical AI innovation for complex operations
Our innovation focus is iCura and its Guardian AI assurance capability. We explore how software and AI can make operational information more useful while preserving human accountability.
Research focus
The questions we investigate
Reducing fragmented information
Can relevant operational information be brought together so teams spend less time searching for it?
Making findings easier to examine
How should AI-assisted findings be presented so people can check the evidence behind them quickly?
Evaluating assistance in real contexts
Does a given form of assistance actually help operational teams, and what does it cost them in review effort?
Our method
From idea to evidence
This is the approach we require for new AI capabilities in iCura.
Step 1: Problem definition
Agree a clearly defined operational problem and who it affects.
Step 2: Prototype
Build the smallest useful version to test the idea.
Step 3: Controlled evaluation
Test with synthetic or approved data against agreed measures.
Step 4: Review
Examine results, errors and limitations with the people who would use it.
Step 5: Release decision
Release, revise or stop — based on the evidence, not momentum.
Research themes
Innovation themes
All of the themes below are at the research stage. None describes a released feature.
Status: research
Information retrieval and context
- Hypothesis
- Showing related records together helps staff understand a situation faster.
- Intended user
- Registered managers and quality teams
- Evaluation approach
- Task-based comparison using synthetic records; accuracy of retrieved context.
- Next milestone
- Define evaluation tasks with operational staff.
Status: research
Workflow assistance
- Hypothesis
- Suggested next steps can reduce missed follow-ups without adding review burden.
- Intended user
- Operations teams
- Evaluation approach
- Usefulness ratings, error rates and time spent reviewing suggestions.
- Next milestone
- Prototype on a narrowly defined workflow.
Status: research
Assurance evidence
- Hypothesis
- Organising evidence around review questions makes assurance work more continuous.
- Intended user
- Quality and assurance teams
- Evaluation approach
- Coverage gaps, incorrect links between evidence and questions, reviewer feedback.
- Next milestone
- Agree evidence categories and review criteria.
Status: research
Evaluation methods
- Hypothesis
- Consistent evaluation measures make AI features safer to compare and release.
- Intended user
- RIAConsult product and engineering teams
- Evaluation approach
- Repeatability of measures across features and datasets.
- Next milestone
- Document a shared evaluation checklist.
Evaluation
Responsible evaluation
When we evaluate an AI capability we look at incorrect outputs, gaps in coverage, whether it is genuinely useful and how much review work it creates for people. Evaluation uses synthetic or approved datasets. Known limitations are recorded, and a capability must pass agreed release gates before it is made available.
Partnerships
Collaborate with RIAConsult
We invite care providers, research collaborators and deployment partners to get in touch. We welcome conversations about clearly defined operational problems and responsible evaluation.
Discuss a collaboration
Tell us about the operational problem you have in mind and how you would like to work together.