Villa Liza · Boksburg · Ekurhuleni pilot

Lights that brighten the moment someone needs them

Auralight Grid dims LED poles to a low idle, then raises output when radar and an edge AI classifier detect a pedestrian or vehicle. Edge microphones flag sudden events (glass-break, repeated impacts) for a human operator to verify — no audio is stored. Each pole reports its own health so faults are logged before anyone calls.

See how it works Explore 3D model
30%
Idle output
<10 min
Target fault detect
50–70%
Energy design target
0 audio
Stored recordings

This solution requires current AI models to operate. Framed as a live testing product in pilot stage.

Why dark or always-on poles cost more than light

In Ekurhuleni, vandalism, copper theft and failed photocells leave routes dark or permanently lit. Faults are often found only after residents complain. Energy is wasted; safety suffers.

1
Pole fails or stays onPhotocell or driver failure / vandalism
2
Route goes darkor burns 24/7 wasting energy
3
Residents complainCall centre / social media
4
Work order createdOften days later
5
Crew dispatchedParts may be out of stock
6
Repair or replaceSame pole can fail again
7
No pattern learningRoot causes repeat
8
Safety & cost cycleDark streets + budget pressure

Energy waste

Poles stuck at 100% or with failed photocells burn through municipal electricity budgets. Adaptive dimming is a design target of 50–70% saving versus constant-on.

Reactive maintenance

Faults are discovered by complaint. Auralight poles self-report driver, sensor and power health so tickets can open before the route goes dark.

Night safety

Dark pedestrian routes increase risk. Light that follows movement improves visibility where people actually walk, without leaving the whole street at full output all night.

Liability note: Auralight Grid provides decision-support signals and health telemetry. Final operational and safety decisions remain with the responsible municipal or contracted operators.

From edge detection to human decision

1. SenseRadar + ambient + mic patterns
→
2. ClassifyEdge AI on pole
→
3. Act locallyRaise / lower light
→
4. ReportHealth + event to gateway
→
5. ScoreConfidence + context
→
6. ExplainOperator view
→
7. Human decideApprove / field check

Scenario Theatre

Five short scenarios that keep validating why adaptive light + acoustic events + human gate matter.

Night walk home — 21:40

Resident enters radar zone. Pole raises from 30% to 100% for a 20 m bubble. After departure, returns to idle. No central command required for the light change.

Glass-break pattern

Edge mic detects glass-break signature. Event packet (time, pole ID, confidence, no audio) sent to control room. Operator verifies and decides dispatch. Zero audio retained.

Failing driver

Pole health telemetry shows rising driver temperature and lumen drop. Work order suggested before the light goes dark. Human approves or modifies the ticket.

Always-on photocell failure

Ambient sensor + schedule logic flags a pole that never dims at dawn. Ticket opened automatically for photocell or controller check.

Repeated impact events

Multiple loud impact patterns on one corridor in one hour. System surfaces a pattern card so an operator can decide whether to escalate to visible policing presence.

Roles required by this product

Roles are derived from the product itself. Models and services are listed as the stack the pilot is designed around; no pricing or cost ranges are shown.

Edge presence classifier

Runs on the pole controller. Distinguishes pedestrian / vehicle / empty from radar and PIR. Keeps light decisions local and fast.

Candidates: lightweight CNN / decision tree on MCU or Jetson-class edge; TensorRT / CUDA where GPU edge is used.

Acoustic event detector

On-device pattern matching for glass-break and repeated impact signatures. Emits event metadata only — never stores or transmits audio.

Candidates: small spectrogram classifiers, specialised audio event models; edge inference only.

Health & anomaly scorer

Combines driver temperature, current draw, lumen proxy and ambient readings to flag poles that will fail soon.

Candidates: time-series models on AWS IoT / SageMaker pipelines; Bedrock for explanation text.

Operator explanation layer

Turns scores and events into plain-language cards for the control room so a human can decide quickly.

Candidates: GPT-family, Claude, Gemini for structured explanation; always reviewed by a human.

Human Reviewer (mandatory)

Every recommendation that creates a work order, dispatch or escalation requires an explicit Approve / Modify / Reject / Field check / Escalate click. No auto-approve.

Cloud helpers

AWS IoT Core for device twin & messaging, S3 for telemetry archives, Lambda for rules, optional SageMaker / Bedrock for cloud-side models. NVIDIA Jetson / TensorRT / NIM for edge where higher compute is needed.

How the models communicate

1. Sensors on the pole produce raw readings.
2. Edge classifiers decide light level and whether an acoustic event is present.
3. Only metadata and health scores leave the pole.
4. Cloud services aggregate, score longer patterns and draft explanations.
5. A human operator sees the card and must act before any work order is created.
6. Decisions are logged for audit.

Interactive pole model

Assembly and operational views. Drag to rotate, scroll to zoom. Use the stress slider to change behaviour and status.

IDLE 30%

The model shows a single street pole with LED head, sensor pod (radar + ambient + edge mic), and status light. In idle state the light is dim (blue status). When a presence event is simulated the light rises and a warm pool appears. Under high stress the driver shows a fault colour and a maintenance ticket state is indicated. Click behaviour is simplified for the pilot: the controls above drive the scene. Proportions and materials aim for a realistic night-street feel.

What changes if the pilot succeeds

Director note. “We started Auralight because dark or always-on poles in places like Villa Liza are both a safety problem and a quiet budget leak. The goal of this pilot is simple: prove that adaptive light plus on-device event detection plus a human decision gate can cut energy, surface faults faster, and give operators a clearer picture — without storing conversations.” — Eirshwin Manogaren Arumagum Moonsamy

Municipal energy teams

Design target 50–70% energy reduction versus constant-on on the pilot corridor. Self-reported health reduces complaint-driven call-outs.

Public safety / EMPD coordination

Acoustic event cards (glass-break, repeated impacts) arrive as metadata only. Operators decide whether to verify or dispatch.

Residents on the corridor

Light where people walk at night. Faster repair of failed poles. No audio recording of conversations.

Why it can work (pilot-stage projections grounded in real SA context)

  • Ekurhuleni manages on the order of 196 000 streetlights; official statements put operational rate around 83%, with vandalism a primary cause of outages.
  • LED retrofit + smart controls programmes (SANEDI / EEPBIP and municipal tenders) already target large energy and maintenance savings.
  • On-device acoustic pattern detection is used in other public-safety deployments; the POPIA-safe design choice is to never store or transmit audio.
  • Human-in-the-loop is non-negotiable for any work order or escalation generated by the system.

Problem context and verified sources

Plain-English context for Villa Liza / Ekurhuleni lighting and safety pressures, with real links. No invented citations.

Ekurhuleni street lighting reality

City statements and local reporting document large inventories, vandalism-driven failures, and ongoing LED retrofit and maintenance programmes. Repair backlogs and always-on poles from damaged photocells are repeatedly noted in Boksburg-area coverage.

City of Ekurhuleni

SANEDI & public lighting efficiency

SANEDI and the EEPBIP programme support energy-performance approaches for public lighting, including smart LED + central management systems and private-finance models for municipal upgrades.

SANEDI

Energy & lighting analysis

IEA and related work on lighting efficiency underpins the design target ranges for adaptive LED systems versus legacy high-pressure sodium.

IEA

POPIA & sensor processing

Acoustic and presence sensors process data that can engage POPIA. Design choice: purpose limitation to safety and maintenance, on-device analysis, no audio storage, public notices, and human review of any action that affects people or infrastructure.

Information Regulator of South Africa

Ten problems and how Auralight Grid answers them

1. Always-on streetlights after photocell failure

Problem: Damaged day-night controls leave poles burning 24/7.

How the solution improves it: Ambient + schedule logic flags never-dimming poles and opens a health ticket.

Models: rule + time-series anomaly; explanation via LLM candidates (GPT / Claude / Gemini).

2. Dark pedestrian routes at night

Problem: Failed or sparse lighting on walking corridors.

How the solution improves it: Radar-triggered raise from idle to full in a local light bubble that follows movement.

Models: edge presence classifier; optional Jetson / TensorRT.

3. Complaint-driven maintenance only

Problem: Faults discovered after residents call.

How the solution improves it: Each pole reports driver, sensor and power health continuously.

Models: health scorer on IoT telemetry; AWS IoT Core + Lambda rules.

4. Copper theft and vandalism of poles

Problem: Infrastructure stripped or knocked down.

How the solution improves it: Impact / tilt and power-loss events surface quickly as metadata for operator review.

Models: edge event detector; human decision gate.

5. Energy budget pressure on municipalities

Problem: Street lighting is a major municipal electricity load.

How the solution improves it: Adaptive dimming design target 50–70% vs constant-on on pilot corridors.

Models: local control loops; cloud aggregation for reporting.

6. Glass-break and sudden impact events

Problem: Vandalism or break-ins along lit corridors go unnoticed until later.

How the solution improves it: On-device acoustic pattern detection sends event cards without storing audio.

Models: spectrogram / event classifiers on edge only.

7. Slow repair cycles and parts shortages

Problem: Work orders lag; depots run out of parts.

How the solution improves it: Earlier fault signals allow planned rather than emergency call-outs.

Models: predictive health scoring; operator prioritisation.

8. Lack of corridor-level pattern view

Problem: Single-pole tickets hide repeated problems on one street.

How the solution improves it: Aggregated event and health views highlight corridors that need a different response.

Models: cloud pattern layer; explanation cards for operators.

9. Privacy risk from always-listening microphones

Problem: Public concern about recording conversations.

How the solution improves it: Analysis stays on-device; only event metadata leaves the pole; no audio retained.

Models: edge-only; POPIA purpose limitation and notices.

10. Operators flooded with raw alarms

Problem: Unfiltered alerts create noise and missed real events.

How the solution improves it: Scored, explained cards with mandatory human decision before work orders.

Models: scoring + LLM explanation candidates; Human Reviewer role.

20 areas — mix of Live pilot and Prospective

Real locations in and around Ekurhuleni / East Rand. Live status indicates pilot telemetry interest; Prospective indicates planned expansion. Live weather feed used as a public API demonstration.

Live coverage demonstration

Public weather for Boksburg / OR Tambo area is fetched live. Pole telemetry in the pilot remains tagged “Pilot test data”.

Loading live weather…

Source: Open-Meteo · timestamp updated on fetch

Auralight Grid by Brightwell

Brightwell builds adaptive public-lighting and edge-safety products for South African municipalities. Auralight Grid is the street-lighting product currently in pilot-stage live testing in Villa Liza, Boksburg.

Legal entity

Brightwell · Registration number 2026/912954/07 · Private Company · In Business · Registered 12 November 2026 · Financial year end: March · Tax number 9030319330

Registered office & postal: 7 Impala Street, Parkdene, Boksburg, Gauteng, 1459

Director: Eirshwin Manogaren Arumagum Moonsamy (appointed 12 November 2026)

Departmental emails: info@brightwellarumagum.co.za · rfq@brightwellarumagum.co.za · admin@brightwellarumagum.co.za · finances@brightwellarumagum.co.za · governmentsolutions@brightwellarumagum.co.za

Director

“We built Auralight so that a resident walking home in Villa Liza gets light when they need it, the municipality spends less on electricity and call-outs, and no one has to worry that a microphone on a pole is storing their conversation. The pilot is how we prove the engineering and the governance together.”

— Eirshwin Manogaren Arumagum Moonsamy, Director, Brightwell

Talk to the pilot team

Direct: info@brightwellarumagum.co.za · governmentsolutions@brightwellarumagum.co.za · rfq@brightwellarumagum.co.za

Pilot testing access

Select a role to enter the Open Pilot Workspace. All telemetry and AI outputs on the next screen are part of the pilot demonstration environment.

PILOT STAGE — LIVE TESTING PRODUCT · All telemetry, predictions and AI outputs on this screen are part of the pilot demonstration environment