Water Intelligence Δ

Where the name comes from.

The practice is older than we tend to remember. In the Heimskringla — Snorre Sturlason's thirteenth-century chronicle of the Norse kings — the land is described through its landmarks: the headlands, the rivers, and the varder. Stone cairns built on high ground to mark passage, to say someone was here, this is the way. Each cairn assembled by human hands, one stone at a time, accumulating into a permanent signal visible across distance and time.

Sturlason wrote of a world navigated by accumulated knowledge — knowledge passed forward, encoded in landscape, made durable. The cairn was not just a marker. It was a record: of who had passed, what they had found safe, and where the next safe ground lay.

A cairn, viewed from the side, resolves into a triangle. The same shape as Δ.

Δ carries three layers of meaning in a single symbol. In mathematics, it is the symbol for change — the difference that matters, the anomaly that demands attention. In hydrology, a delta is where flows converge and are understood, where a network meets the wider world. And the cairn: the ancient waymarker built stone by stone by those who came before, so that those who follow know where they are and where to go next.

That practice is unbroken. Today, children still stack stones on Norwegian summits — each adding their stone to what was already there, contributing to a shared structure they did not start and will not finish. Not because anyone told them to. Because the logic is universal: you take what you found, you add what you know, and you leave it better than you arrived.

Δ is the digital expression of the same logic. Your network already holds the knowledge — in telemetry, in risk scores, in decades of operational data. Δ assembles it, makes it speakable, and passes it forward — stone by stone, answer by answer — so the next person always knows where they stand.

What Δ does.

Δ is an AI assistant built on top of PipeFusion's graph neural network model of your water or wastewater network. It translates the structural knowledge locked inside that model — topology, flow estimates, risk scores, telemetry — into direct, conversational answers.

Ask about a pressure zone, a failing asset, a catchment with elevated infiltration, or the consequence of a planned intervention. Δ returns a reasoned answer grounded in your actual network data, not generic hydraulic knowledge.

What Δ draws on.

Δ's answers are composed from multiple data layers that PipeFusion already holds. No new data collection is required — the LLM is a reasoning layer on top of existing models.

Δ receives a structured context object containing current values from each layer before generating a response. It does not call external APIs or retrieve web content. All reasoning is bounded by the data InfoTiles already holds.

What Δ can do today.

Δ's scope is deliberately bounded to areas where InfoTiles data provides genuine depth. Capabilities outside this scope are not surfaced.

Current scope tags:

Ask Δ a question.

The example below is an actual prompt run against a utility network. Δ combines SewerIntelligence I&I telemetry with PipeFusion risk scores to generate a prioritised CCTV inspection plan — structured, ranked, and ready to act on.

Live query · SewerIntelligence I&I telemetry + PipeFusion LoF/CoF risk scores

What users say about Δ.

How Δ is built.

Δ uses a retrieval-augmented generation (RAG) approach. At query time, relevant segments from the network model, risk scores, and reference documentation are retrieved and injected into the LLM context. The model does not retain state between sessions and is not fine-tuned on customer data.

Infrastructure: Microsoft Azure AI Foundry — EU-hosted, stateless inference, no training on customer data. The LLM is replaceable; the retrieval and context construction pipeline is where the domain specificity lives.

Human-in-the-loop is embedded at the formal recommendation layer. Δ surfaces evidence and methodology; decisions requiring sign-off remain with the operator or asset manager. This is a design principle, not a disclaimer.

What Δ is not built for.

What makes Δ distinctive.