Technology

Four layers.
All of them ours.

Most companies in this space build one layer and rent the rest. We took the heavier route, because the value here isn't in a single tool — it's in several products reading one set of data and one judgement.

Owning the lower layers is why adding a new product costs almost nothing to onboard, and why results across products can be measured on the same basis.
L1Infrastructure

Built and run by us

Compute, storage and end-to-end observability supporting every product on one operational footing, with cost attributed per product and per tenant. Hosted and maintained in Australia. You get the capability without taking on the infrastructure.

L2Model

VODA

Our own domain-specific model, trained for how AI engines decide which brands to cite. It handles query-intent modelling, visibility scoring, citation attribution and the generation of fixes that can actually be deployed. It is not a prompt layer wrapped around a general-purpose model — that distinction is the reason the judgement holds up on questions a generic assistant gets wrong.

L3Control & brand data

One record, and permission to touch it

Two things live here. A governed brand record — facts held to a single correct answer, messaging that can carry variants, and rules that bind output — read by every product you enable. And a control layer that scopes what any agent may do: no standing access, every action running on a short-lived, cryptographically signed permission limited to that task, written to an audit log you can read and export. Access is granted per product rather than pooled centrally, so a distributor or agency can be given a restricted view without seeing your whole record.

L4Products

The Lumi line

LumiGEO is live; LumiCast, LumiReach and LumiRadar follow through 2026. They aren't four parallel tools — they work off the same record and the same control layer, which is what lets an effect measured in one be attributed across all of them.

The rule layer

Blocked at generation,
not caught in review.

Automated deployment is only safe if something stops the wrong thing being written in the first place. For companies selling into markets they don't live in, that's the difference between a useful system and a liability.

Claims are market-specific

Permitted and prohibited wording is maintained per market, side by side. A claim valid at home may be meaningless or non-compliant where you're selling.

Certificates carry expiry

A claim can only cite a certificate that's uploaded and in date. When one lapses, the wording that depended on it retires itself.

Parameters are never inferred

Load ratings, materials, dosages and safety figures are quoted only from values you confirmed in writing. The system does not generate or extrapolate them.

The rules bind every path. Your own team, your distributors, your agency and our agents all write through the same layer. There is no route around it — which is what makes it worth maintaining.

Original research

AI engines don't
copy each other's answers.

This finding sits underneath every product decision we've made, and it's why we cover every major engine instead of building one good demo.

20–40%Overlap between the brands different engines name when asked the same question. Optimise for one and you've given up the rest.
WeakCorrelation between traditional search ranking and AI citation rate. The two run on different logic and have to be tracked separately.
It decaysEngine logic shifts, competitors keep publishing, content ages. Visibility erodes on its own if the work stops.
Same question · seven enginesBrands named
Engine A52%
Engine B71%
Engine C24%
Engine D44%
Engine E18%
Engine F37%
Engine G12%
Overlap across this set: 31%. Illustrative of the pattern we track continuously.

Edisyn ongoing tracking research · figures are indicative and vary by category

Next

Edisyn Dashboard

What the stack looks like from the customer's side: modules, autonomy levels and plans.

Or

The Lumi product line

LumiGEO today, and what lands through 2026.