Energy & Utility Network Intelligence Quotient

Intelligence for
the distribution
network.

An enterprise intelligence platform for energy and utility networks. EUNIQ unifies meter, network, billing and service data into one topology-aware model — then forecasts demand, detects anomalies, scores risk, and tells a field crew what to act on first.

One intelligence layer · Electricity, gas and water · Any utility, any market

01
Where is the loss?
Loss, theft and non-revenue intelligence across the network
02
What is about to fail?
Demand forecasting, anomaly detection and asset risk scoring
03
What should the crew do today?
Ranked, evidence-backed work a field team can act on
Designed for
Worldwide
Utility networks share one structure — wherever they are
Every distribution network runs the same hierarchy — substation, feeder, transformer, consumer. EUNIQ is built against that structure, not against one country's systems.
Where EUNIQ applies

One architecture, every utility network

Utilities worldwide face the same three questions: where is the loss, what is about to fail, and what should the crew do today. The regulator differs, the vocabulary differs, the physical structure does not. EUNIQ is built against that structure.

UNIVERSAL STRUCTURE

The hierarchy is the same everywhere

Substation → feeder → transformer → consumer for electricity; the equivalent zone → main → district → connection for gas and water. The data model is built against topology, so a new market is a configuration exercise, not a rebuild.

GLOBAL STANDARDS

Built to international standards

CIM IEC 61968 / 61970 for network and asset data, DLMS / COSEM for metering, MQTT and OPC-UA for telemetry, open REST and event APIs throughout. Standards written for utilities worldwide, not one vendor's stack.

SHARED PRESSURE

The same problem, different names

Non-technical loss, AT&C loss, non-revenue water, unaccounted-for gas — four names for one question. Add ageing assets, electrification, distributed generation and demand volatility, and every utility is being asked to see further with the data it already holds.

Available
EUNIQ Grid — electricity
Loss and theft analytics, outage and feeder / transformer intelligence, demand forecasting and network resilience for distribution utilities.
Available
EUNIQ Gas — gas networks
Leak and pressure intelligence, demand forecasting, pipeline and asset anomaly detection, and unaccounted-for gas analytics.
Available
EUNIQ Water — water networks
Non-revenue water, leakage and pressure intelligence, demand forecasting and asset risk across the distribution network.
Built for
Partly instrumented networks
Utilities where smart metering is mid-rollout, asset registers are incomplete and source systems disagree. Most platforms assume a clean, fully instrumented network. EUNIQ is designed for the one you actually have.
Deployment
Cloud, on-premise or sovereign
Cloud-native by design and deployable inside a utility's own estate where regulation, data residency or critical-infrastructure policy requires it.

Every layer is available for deployment. The platform is deliberately engineered against the hardest version of the problem — incomplete, inconsistent, partially instrumented networks — rather than the easiest.

How EUNIQ works

From a raw meter reading to an action a field officer can take

Five stages, in order. The value isn't in any single model — it's that every stage operates over one shared, topology-aware representation of the network.

01

Ingest and unify

Feeder and DT interval readings, asset master, network topology, billing, outage and grievance records are aligned to a common time base. Every record is placed within the substation → feeder → DT hierarchy.

02

Resolve identity

Billing, GIS and metering systems disagree about which consumer sits on which transformer. EUNIQ reconciles those conflicting mappings into one queryable graph — the step that makes everything downstream possible.

03

Forecast

An LSTM/GRU sequence model learns previous-hour, day and week load, daily shape, weekday and weekend behaviour, peak periods and seasonal trend, then predicts expected load and peak 24–72 hours ahead — conditioned on the network hierarchy, not treated as a flat time series.

04

Detect and attribute

An LSTM autoencoder learns each asset's normal behaviour and flags deviation by reconstruction error — unsupervised, because confirmed failure labels are scarce in any utility. Residuals are correlated across the hierarchy to separate a technical condition from a commercial one.

05

Act, then learn

Every alert carries the signals that produced it, in a form a field officer can act on. What the engineer actually finds is recorded and becomes ground truth — so precision improves with each cycle, on the utility's own network.

EUNIQ Platform

One intelligence platform. Every utility.

One intelligence layer, multiple utility domains. Every module draws on the same data foundation and the same pattern engine — which is why adding a domain is configuration, not a rebuild.

Pattern Engine Data Fabric Knowledge Graph EUNIQ Grid EUNIQ Gas EUNIQ Water EUNIQ Cockpit EUNIQ Command EUNIQ Field EUNIQ Enterprise energy and utility intelligence platform CONSUMPTION LAYER — COMMON TO EVERY DOMAIN DOMAIN MODULES — ALL AVAILABLE FOR DEPLOYMENT
CONSUMPTION LAYER — common to every domain
EUNIQ Cockpit
Dashboards and a composite intelligence score for loss, resilience, service and asset risk.
EUNIQ Command
Zonal, regional and executive operations centre views.
EUNIQ Field
Inspection, evidence, work orders, maintenance and outcome feedback.
DOMAIN MODULES All available
EUNIQ Grid
Technical & non-technical loss, theft, outage, feeder and transformer intelligence, demand, resilience.
EUNIQ Gas
Leak, pressure, demand, pipeline and anomaly intelligence.
EUNIQ Water
Non-revenue water, leakage, pressure, demand, asset intelligence.
PLATFORM FOUNDATION — built once, shared by every module
EUNIQ Pattern Engine
Correlation, anomaly detection, forecasting and risk scoring — the intelligence layer every domain module runs on.
Unified Data Fabric
Ingestion, alignment and quality across every source system.
Utility Knowledge Graph
Topology-aware graph of assets, connections and consumers.
SEEUNDERSTANDPREDICT DECIDEACT
Consumption layer — common to every domain
EUNIQ Cockpit
Dashboards and a composite intelligence score for loss, resilience, service and asset risk.
EUNIQ Command
Zonal, regional and executive operations centre views.
EUNIQ Field
Inspection, evidence, work orders, maintenance and outcome feedback.
Domain modules — all available for deployment
EUNIQ Grid Available
Technical & non-technical loss, theft, outage, feeder and transformer intelligence, demand, resilience.
EUNIQ Gas Available
Leak, pressure, demand, pipeline and anomaly intelligence.
EUNIQ Water Available
Non-revenue water, leakage, pressure, demand, asset intelligence.
Platform foundation — built once, shared by every module
EUNIQ
Enterprise energy and utility intelligence platform.

EUNIQ Pattern Engine

Correlation, anomaly detection, forecasting and risk scoring — the intelligence layer every domain module runs on.

Unified Data Fabric
Ingestion, alignment and quality across every source system.
Utility Knowledge Graph
Topology-aware graph of assets, connections and consumers.
Utility Knowledge Graph

Your network, as a model a machine can reason over

Most platforms treat a meter reading as a number in a table. EUNIQ places every reading at a known point in a known network — so a signal can be interpreted in the context of the asset above it, and an anomaly can be attributed to a cause.

Substation Feeder Transformer Consumer Case
SELECTED NODE
Why a graph and not a table

Billing, GIS and metering systems disagree about which consumer sits on which transformer. Resolving those conflicts into one queryable structure is what makes loss attribution, topology-aware forecasting and evidence-backed cases possible.

See it live in the cockpit ↗
Technology

What EUNIQ is built on

A utility-domain algorithm library, deep learning, graph methods and agentic AI on one engineering core — assembled by a team that has shipped production systems at scale, and reviewed by doctoral advisors before anything is built.

30+
AI agents in the framework
12
Model families in the stack
Utility-native
Algorithms built for the network, not adapted to it
PhD
Led model & methodology review
Utility algorithm library

Domain algorithms, not generic analytics

Machine learning alone does not understand a network. EUNIQ pairs it with the engineering methods utilities already trust — power-systems, hydraulic and gas-network analysis — so a model's output can be reconciled against physics rather than accepted on faith. This is the layer most analytics platforms do not have.

ELECTRICITY
Distribution state estimation Load flow analysis Technical vs non-technical loss segregation Energy balance & audit Transformer thermal ageing Phase imbalance detection Voltage profile analysis Coincidence & diversity factors Tamper-event correlation Outage cause classification
GAS
Pressure-drop leak localisation Unaccounted-for gas balance Pipeline integrity scoring Flow & demand profiling
WATER
Minimum night flow analysis District metered area balance Hydraulic model calibration Burst detection & localisation
CROSS-DOMAIN
Topology reconstruction Consumer–asset identity resolution Asset health index scoring Load & demand profiling Peak & capacity headroom Meter data validation, estimation & editing Interval gap reconstruction Revenue protection scoring
Artificial intelligence & machine learning
DEEP LEARNING

Sequence & representation models

LSTMGRU Seq2SeqTemporal Fusion Transformer N-BEATSInformer TCNAttention

Load carries structure across time. Recurrent and attention architectures learn that sequence directly instead of approximating it through hand-built lag features.

ANOMALY & UNSUPERVISED

Detection without labels

LSTM AutoencoderVariational AE Isolation ForestDBSCAN One-Class SVMMatrix Profile Change-point detection

Confirmed, dated transformer-failure records are scarce in every utility, in every market. Unsupervised models learn normal behaviour and score deviation, so no failure labels are required to start.

GRAPH & TOPOLOGY

The network as a first-class object

Graph Neural NetworksGraphSAGE GATNode2Vec Entity resolutionProbabilistic matching

Substation → feeder → DT is modelled as a graph, not metadata. Signals are read in the context of the level above them, and conflicting consumer mappings are resolved with a confidence score on every link.

CLASSICAL ML & STATISTICS

Baselines that keep us honest

XGBoostLightGBM CatBoostRandom Forest SARIMAProphet Kalman filterQuantile regression

Every deep model is measured against a simpler baseline. If a gradient-boosted tree wins on an asset class, we ship the tree.

AGENTIC AI & LLM

30+ specialised agents

LangGraphTool-calling agents RAGVector search pgvectorQdrant Transformer NLPMultilingual

Narrow, independently testable agents for ingestion, identity, forecasting, anomaly, attribution and feedback — orchestrated rather than fused into one opaque model.

EXPLAINABILITY

Evidence, not verdicts

SHAPLIME CaptumCounterfactuals Attention attributionConfidence scoring

Every alert ships with the signals that produced it and the number behind each. Factors are reported as probable — a confirmed cause needs field validation only the utility can supply.

Platform & engineering stack
Python core
Python 3.12pandas NumPyPolars SciPyscikit-learn statsmodelsNumba
ML frameworks
PyTorchTensorFlow KerasPyTorch Geometric DartsGluonTS NixtlaPyOD
Data engineering
Apache SparkKafka Airflowdbt DuckDBApache Arrow Great Expectations
Storage
PostgreSQLTimescaleDB ClickHouseDelta Lake ParquetNeo4j RedisS3 / MinIO
MLOps
MLflowDVC OptunaRay FeastONNX EvidentlyBentoML
Services & API
FastAPIgRPC CeleryPydantic OpenAPIWebSocket
Cockpit & visualisation
ReactTypeScript D3ECharts deck.glMapLibre
Infrastructure
DockerKubernetes TerraformGitHub Actions PrometheusGrafana On-prem or cloud
Utility interoperability
CIM IEC 61968 / 61970 DLMS / COSEMMQTT ModbusOPC-UA MDM / HESGIS / ESRI
Security & governance
OAuth2 / OIDCRBAC AES-256 at restTLS 1.3 Audit loggingAnonymisation GDPR-alignedData residency
DEPTH BEHIND THE STACK

Algorithm experience, and independent review

Our founding team has delivered mission-critical systems at scale — real-time credit authorisation platforms, programmes of 2,500 engineers, and an agile engineering foundation for 30,000 people. That is where the discipline behind this platform comes from: correctness under load, reproducibility, and systems that are not allowed to fail.

Model selection and methodology are reviewed by doctoral advisors on our team, alongside subject-matter advisors with distribution-sector operating experience. We also work with university research groups — sponsored problem statements and independent validation of our models against instrumented network data.

PhD
Doctoral advisors Model selection, methodology and research direction reviewed before build
SME
Utility practitioners Distribution-sector operating experience, so models reflect how a utility actually runs
UNIV
University collaboration Sponsored research and independent validation of model performance
TEST
Held-out validation Trained on one period, measured on a later period the model never sees
Agent framework

30+ specialised agents behind one interface

Rather than one monolithic model, EUNIQ runs a library of narrow agents — each responsible for a single well-defined task, each independently testable. An orchestration layer routes work between them and assembles the result.

30+
SPECIALISED AGENTS
6
AGENT FAMILIES
1
SHARED FOUNDATION
24×7
CONTINUOUS EVAL

Data agents

Ingestion, schema inference, gap detection, unit reconciliation, timestamp normalisation, quality scoring.

Identity agents

Consumer–meter–transformer matching, conflict resolution across source systems, topology validation, orphan detection.

Forecasting agents

Per-asset model selection, feature construction, retraining triggers, drift monitoring, accuracy reporting.

Anomaly agents

Reconstruction scoring, threshold calibration, hierarchical correlation, false-positive suppression.

Attribution agents

Contributing-factor ranking, evidence assembly, confidence estimation, case generation for field teams.

Feedback agents

Inspection-outcome capture, label construction, retraining scheduling, precision tracking over time.

EUNIQ Cockpit

The same data, shaped for whoever is looking

Executives see the score and the trend. Zonal engineers see the assets. Field officers see one job and a form.

EXECUTIVE

Score and trend

Composite intelligence score across loss, resilience, service and asset risk. Zone comparison, and the movement behind each number.

ZONAL

Asset watchlist

Ranked assets, forecast against capacity, open alerts, and a feeder-level view of where a condition is concentrated.

FIELD

One job, one form

The asset, the evidence behind the alert, and a 30-second outcome form that works offline — the mechanism by which the models improve.

See the feeder-level cockpit

A worked example of the distribution transformer drill-down, built on representative feeder data.

Login to EUNIQ → Request an advisory call
EUNIQ Cockpit · Reference view · Feeder
North Region ▸ Substation 07 ▸ Feeder F-11 ▸ DT-024
LIVE · --:--:-- · stream lag 0.8s
18
FEEDERS MONITORED
1,284
TRANSFORMERS
7
OPEN ALERTS
3
HIGH RISK
96.4%
INTERVAL COMPLETENESS
0
READINGS / MIN
DT-024
Sector 4 · 100 kVA · Feeder F-11 · Commissioned 2019
HIGH RISK
Predicted to exceed rated capacity in ~18 hours
Data confidence: Moderate — 19 months history, 4% intervals missing

Utility Knowledge Graph

Live topology around DT-024 — click a node to inspect. Animated paths show data moving through the model.

Substation Feeder Transformer Consumer Case
MATCH (s:Substation)-[:FEEDS]->(f:Feeder)-[:SUPPLIES]->(d:Transformer) WHERE d.id = 'DT-024' RETURN s, f, d

Load and forecast · % of rated capacity

7 days actual · 72 hours forecast
Actual Forecast Confidence band Rated capacity Utility threshold (85%)

Event stream

LAST 60 MIN
PHASE BALANCE
R
Y
B
Imbalance · threshold 10%
VOLTAGE · LAST 60 MIN
Now
Statutory band207 – 253 V
LOADING · LAST 60 MIN
Now
Rated100 kVA

Probable contributing factors

Sustained load growth+18% over 90 days
Peak concentration68% of load in 18:00–22:00
Capacity constraintPredicted peak 104% of rating
Abnormal consumption patternDeviation from 19-month profile
These are probable contributing factors, not a confirmed cause. Field investigation confirms the actual cause.

Deviation from normal behaviour

Reconstruction error against this asset's learned normal band. Deviation began 22 July 2026.

Anomaly score
0.82 / 1.0
Days in deviation
23
Rank on feeder
3 / 42

This transformer in its feeder

7 of 42 transformers on Feeder F-11 are flagged — this may be a feeder-level condition rather than an asset-level one.

DT-017112% · HIGH
DT-031106% · HIGH
DT-024 (current)104% · HIGH
DT-00891% · MEDIUM
DT-04288% · MEDIUM
AGENTS initialising…

Representative data shown for demonstration only. Figures are illustrative and do not depict any real network or customer.