Agentic data analytics · Sydney, Australia
Your systems, connected into one place, with agents on top. The question you would put to an analyst, answered from your own data, with the query and the sources behind every number.
Which of our packaging costs went up more than ten per cent this year?
Four lines, and cartons moved the most, up fourteen per cent.
Somebody asks how the quarter is going. The answer takes a week, arrives as a spreadsheet, and two people produce two different numbers from the same systems. Nobody is lying and nobody is lazy. The numbers live in whatever the business has collected over the years, and nothing reconciles them. The decisions that wait a week for that answer are the expensive ones.
The questions this answers without a queue
It runs on your own infrastructure, and each layer does one job.
A chat window on your own network. Your choice of model, from a frontier model with your own key to one running inside the building.
LibreChat
A warehouse built to answer questions, with an agent interface in front of it. The agent writes the query, and the answer comes back as a table or a chart.
ClickHouse with MCP
Every prompt, tool call, cost and latency traced and scored, so an answer can be audited rather than taken on trust.
Langfuse
Anyone can install the layers. Connecting your data is the work.
Your numbers are not sitting in a warehouse waiting to be asked. They are in a point of sale database, an accounting file, a shelf of spreadsheets and somebody's inbox. I connect those systems, match them on the same keys, reconcile every total against the source, and put the agents on top. That is the work nobody quotes for, and it is what makes the answers worth having.
Your systems
The pipe
The warehouse
One place every number comes from, and one place to ask questions of.
The answers
Tables, charts and briefs, each traced back to the rows behind it.
Three demonstrations on synthetic data: the question as a person asks it, the paperwork as it arrives, and two sets of numbers that never agreed.
Illustrative example. Synthetic data.
The question, and the answer
Which of our packaging costs went up more than ten per cent this year?
Four lines, and cartons moved the most, up fourteen per cent.
The answer arrives as a chart and a table. No dashboard to build, and no analyst in the queue.
The query, and where the numbers came from
SELECT supplier,
unit,
round((max(price) - min(price)) / min(price), 3) AS rise
FROM purchasing.price_history
WHERE year = 2026
GROUP BY supplier, unit
HAVING rise > 0.10
ORDER BY rise DESCWritten by the agent, run against your warehouse. The prompt, the query, the rows and the cost of the answer are all recorded, so any figure in this page can be checked later.
Illustrative example. Synthetic data.
| Shipment | Containers | ETA | Status |
|---|---|---|---|
| Shipment 014 | 1 x 40'HQ | 14 Mar | On time |
| Shipment 013 | 2 x 40'HQ | 2 Mar | Cleared |
| Shipment 011 | 1 x 20'GP | 9 Feb | Delivered |
| Freight exposure, quarter | USD 38,600 | Fixed | Sourced |
Every number traceable, every gap flagged, never guessed.
Illustrative example. Synthetic data.
| Component | Amount | Source | Status |
|---|---|---|---|
| Card settlements, net of fees | 368,025 | Merchant statement | Matched |
| Cash banked | 38,120 | Bank feed | Matched |
| Merchant fees | 535 | Merchant statement | Explained |
| Cash counted, not banked | 3,080 | Till count, 31 March | Explained |
| Net takings | 409,760 | Point of sale | Reconciled |
A variance that looks alarming is four lines and three sources.
Supply chain · Global franchisor
Bills of lading, delay notices, quotes: hundreds of shipping documents reconciled into one auditable picture. Arrivals by port and origin, transit times, freight exposure, every number traceable to a source.
Outcome: control over a container programme nobody could previously see.
Systems · Global franchisor
Ordering, pricing, rosters and reporting had grown into scripts, spreadsheets and half-finished migrations, each one fragile and none of them owned end to end. The pieces were consolidated into a single system.
Outcome: one system, every change reversible, every number traceable.
Three steps, in the order this work happens. You can stop after any of them.
A data audit. What sources exist, who owns each one, where the numbers disagree and by how much. Days, not months, with every number sourced.
The pipe and the stack. Data arriving on its own, agents on top of it, all of it on your infrastructure.
Running it. New sources, broken feeds, reconciliation reports, and a response time we agreed in advance.
Agreed before any work starts
Which source systems are in scope. How often the data arrives. How quickly I respond when something breaks. Everything else is deliberately out of scope, in writing.
Tell me what you have and what you cannot see. Days later you will know what your data can answer, and what it would take.
The problem is with me now. I'll be in touch.