Case study · Climate data · Australia
The climate agent that will not answer from memory.
A nonpartisan climate-data publisher serves a free public site and a paid product used by media houses, researchers and analysts. We built two agents that answer questions about that data in plain English, grounded in a live query every time and cited like everything else on the site.
- The problem
- Every figure sat behind pages chosen in advance, so one custom report started from scratch with an analyst.
- What we built
- Two agents on a dedicated data layer: an everyday agent and a deep-research agent.
- The result
- A report that took one to two days now takes minutes, in the publisher's voice, with every figure cited.
- Status
- Live in production since 2025.
The same problem shows up in
- Research desks
- Regulators
- Investment committees
- Compliance teams
- Clinical registries
If your organisation can be held to its numbers, your agent has to be traceable to a source, not merely fluent.
Before and after.
Every figure was reachable only through a page somebody had decided on in advance, so a report on one topic started from zero.
- 1Work out which pages hold the answer
- 2Pull the figures out of each dataset
- 3Combine them by hand
- 4Cross-check every number
- 5Chart it and lay it out to brand
- 6Write it up in the house voice
One to two days of an analyst's time, or more, for one report on one topic.
The question goes in the way an analyst would say it out loud.
- 1Ask in plain English
- 2Report, charts and citations
No longer done by hand
- Pull the figures out of each dataset
- Combine them by hand
- Cross-check every number
- Chart it and lay it out to brand
The report comes back in minutes, in their voice, with every figure cited.
- 190
- Data tools
- 2
- Agents live
- 100%
- Figures cited
- 3 days
- Discovery
One per published view, so it cannot ask what the site cannot answer.
An everyday agent and a deep-research agent.
Named source, down to the slice of the dataset.
Then a pilot in under three weeks.
Why the bar was set so high.
The data is published under an open licence, so a wrong number does not wait to be corrected. It travels.
Traceable
Named source, down to the slice of the dataset the figure came from.
Queried, not recalled
Produced by a database query, never from the model's own memory.
In-house voice
Plain, neutral language with no judgement and no filler.
Honest about limits
Able to say “we don't hold that” rather than approximate.
How we worked.
3 days
Discovery
Learned the domain, how readers already use charts and tables, and how the data is stored.
Under 3 weeks
A narrow pilot
One agent, two subject areas, the real database and the site's own visualisations.
Their call
They decided it was worth it
The pilot was built to be assessed, not admired, and answered real questions on their own data.
Ongoing
Production
Two agents live on a dedicated data layer.
What it does now.
A data layer built to be read
A server in front of the database hands back only what the question needs.
No mental arithmetic
Every sum, even a monthly total, goes through a calculator call.
Cited to the slice
Not just the source name, but the dataset and the filter behind the figure.
No open web
Search, uploads and code execution are all switched off.
Shows its working
Readers see the query that produced the number by default.
Charts chosen by rule
The shape of the data picks the chart, never the model's preference.
Start with one process.
Tell us which part of your work takes the most time. We'll show you what an AI co-worker could take off your team's plate.
Contact us
We'll use your details to reply to your enquiry. See our Privacy Policy.
What to expect
We start with one process and put something real in front of your team within weeks, so you can judge it on your own work.
Thor Signia
How we start
