AI Intelligence Assistant Research track
Ask in plain language. Get answers you can verify.
The Epinova AI Intelligence Assistant is a research-stage capability that would let authorized users query surveillance and programme data in natural language - with every answer grounded in, and cited to, the underlying sources.
The problem
The answer exists - somewhere in the data.
Decision-makers often need a quick, specific answer: which districts have rising positivity, what the forecast says for next month, what was decided at the last partner meeting. Getting it usually means waiting for an analyst. Generic AI chat tools, meanwhile, cannot be trusted with programme data and may produce confident but unsupported answers.
- Simple questions require analyst time
- Generic AI tools are not grounded in authorized data
- Unsupported answers carry real operational risk
Product overview
Grounded, cited and permission-aware.
The assistant is being researched as a retrieval-grounded interface: it would only answer from data the user is authorized to access, cite the records and documents behind each statement, and say clearly when it does not know.
It is not yet available. We are prioritizing evaluation of factual accuracy, citation quality and refusal behaviour before any release.
Key capabilities
What it is designed to do
Natural-language questions
Plain-language queries over surveillance, forecast and programme data.
Citations on every answer
Each statement linked to its source record or document.
Permission-aware retrieval
Answers limited to data the user is authorized to see.
Briefing drafts
Draft situation summaries for human review.
Interactive visualization
Explore AI Intelligence Assistant
Illustrative concept - simulated question, data and citations.
Three fictional areas show positivity rising for three or more consecutive weeks: Kestrel Marsh, Granite Bay and Dune Hollow[1]. Kestrel Marsh has the steepest increase and below-target net coverage[2].
- Weekly surveillance extract, weeks 22–26 (simulated)
- Coverage survey summary, Q2 (simulated)
Research-stage concept - simulated question, answer and sources. Not an available feature.
Example workflow
How the work flows
- 01
Ask
A user asks a question in plain language.
- 02
Retrieve
Relevant authorized records, forecasts and documents are retrieved.
- 03
Answer
A concise answer is drafted with citations to each source.
- 04
Verify
The user can open every source; uncertain answers are flagged.
Inputs and outputs
What goes in, what comes out
Inputs
- Authorized surveillance data
- Forecasts and risk maps
- Programme documents and decisions
Outputs
- Cited answers
- Draft briefings
- Links to source records
Intended users
Who it is for
Senior decision-makers
Fast, trustworthy answers without a data request.
Analysts
Less time on routine queries.
Integrations
Works with the systems you already use
Integration scope is agreed with each partner; connectors marked “planned” are on the roadmap.
- All Epinova modules (planned)
- Document repositories (planned)
Grounding required
No answer without a supporting source.
Says “I don’t know”
Designed to decline when evidence is insufficient.
Evaluation before release
Accuracy and safety testing precede any availability.
Interested in shaping the assistant?
We welcome research and programme partners to help define evaluation criteria.