Lydr AI — real-time video intelligence for construction sites
Lydr AI ziyan@lydrai.com
Real-time video intelligence for construction sites Stage: in development Sydney, Australia

Lydr AI turns the cameras you already have into a live feed of safety, compliance and throughput events — on-site, in real time.

ziyan@lydrai.com→ See the console mock-up ↓
Product
Real-time video-intelligence platform
For
Operations, safety (WHS) and site managers on construction sites
Stage
In development
Company
Lydr Pty Ltd, Sydney, Australia
Background: simulated scene, not live footage.
PPE missing Exclusion-zone entry Near-miss Dwell time Throughput counts
01 — What it doesProduct

A real-time video-intelligence platform for physical worksites.

Lydr AI turns the live camera and video feeds a site already has into a structured, real-time stream of operational events, so operations and safety teams see what is happening across a site without watching screens.

Lydr AI is in development. These five capabilities describe the platform as it is being built.

01

Detect & track

Lydr AI recognises people, equipment, vehicles and zones from existing site cameras in real time.

02

Event & alert stream

Lydr AI raises structured events: PPE missing, exclusion-zone entry, near-miss, dwell time, throughput counts — not raw footage to re-watch.

03

Runs on-site (edge)

Inference runs locally on edge hardware at the site; low latency, and video never has to leave the premises.

04

Fits existing tools

Lydr AI pushes events to the dashboards, radios and systems the team already uses, via a simple API.

05

Tunes to each site

Vision models are fine-tuned to each site's cameras, layout and conditions so detection holds up in the real world.

How it worksCamera feed in, events out
On site: video never has to leave the premises
Step 1

Existing cameras

The live camera and video feeds a site already has.

Step 2

Lydr AI, on-site

Detects and tracks people, equipment, vehicles and zones. Inference runs locally on edge hardware at the site.

Step 3

Structured events

PPE missing, exclusion-zone entry, near-miss, dwell time, throughput counts.

Step 4

Existing tools

The dashboards, radios and systems the team already uses, via a simple API.

02 — Fig. 1, console mock-upSimulated

One site, four views, one event stream.

This is an illustrative mock-up. The scene, the tracks and the events are simulated in your browser. It is not live footage and not a running deployment. Lydr AI is in development.

90 s agoEvent timelineNow
Event streamClick a row
Event payloadIllustrative format
{{ payload }}
PPE missing
{{ n0 }}
Zone entry
{{ n1 }}
Near-miss
{{ n2 }}
Dwell time
{{ n3 }}
Throughput
{{ n4 }}
Fig. 1. Illustrative mock-up of the Lydr AI console. Simulated scene, simulated counts, illustrative event format. The API is in development and the format may change.
03 — Event typesSafety / compliance / throughput

Five events, not hours of footage.

Each event is a structured record with a type, a camera, a time and the objects involved. The animations are illustrations of each event type.

ppe.missingCompliance

PPE missing

A person is detected without required protective equipment, such as a hard hat.

exclusion_zone.entrySafety

Exclusion-zone entry

A person or vehicle crosses into a zone marked as off-limits.

near_missSafety

Near-miss

A person and moving equipment come closer than a set distance.

dwell.timeThroughput

Dwell time

A person or vehicle stays in one area longer than a set time.

throughput.countThroughput

Throughput counts

Vehicles and people passing a gate or a line are counted.

04 — Who it's forConstruction

For operations, safety (WHS) and site managers on construction sites.

Safety (WHS) managers

PPE missing, exclusion-zone entry and near-miss events as they happen.

Operations managers

Dwell time and throughput counts across gates and work areas.

Site managers

One live stream of events across the whole site, without watching screens.

One situation

A lift is under way and a worker walks into the crane exclusion zone.

00:00

An existing site camera sees the worker cross the zone boundary.

Real time

Lydr AI raises an exclusion-zone entry event and pushes it to the tools the team already uses.

Decision

The site manager decides whether to stop the lift and clear the zone, while it is happening.

05 — NVIDIA technologyToday / in development / planned

NVIDIA technology

What Lydr AI runs on NVIDIA technology today, what is in development, and what is planned. Each item names the capability it powers.

In use today

NVIDIA H100 GPU

Lydr AI's vision models are trained on NVIDIA H100 GPUs.

Powers Detect & track · Tunes to each site
Why Training vision models on site video is GPU-intensive.
In development

Detection and tracking models

The detection and tracking models are in training on NVIDIA H100 now. No model is deployed at a customer site yet.

Powers Detect & track
Planned

NVIDIA TensorRT

We plan to optimise the trained models with NVIDIA TensorRT for low-latency inference.

Powers Runs on-site (edge)

NVIDIA DeepStream

We plan to build the multi-camera video pipeline on NVIDIA DeepStream.

Powers Detect & track · Event & alert stream
Capability by NVIDIA component
Capability
NVIDIA H100
NVIDIA TensorRT
NVIDIA DeepStream
Detect & track
In use today
None
Planned
Event & alert stream
None
None
Planned
Runs on-site (edge)
None
Planned
None
Fits existing tools
None
None
None
Tunes to each site
In use today
None
None
06 — StageAs of November 2026

In
development

Lydr AI is in development. No version is deployed at a customer site and there is no public demo yet. The only product visual on this page, Fig. 1, is an illustrative mock-up.

July 2026 Lydr Pty Ltd incorporated in Sydney, New South Wales.
November 2026 Now: training the detection and tracking models on NVIDIA H100.
Next Planned: model optimisation with NVIDIA TensorRT and a multi-camera video pipeline on NVIDIA DeepStream.
Exists today
  • Lydr Pty Ltd, a registered Australian company.
  • Detection and tracking models in training on NVIDIA H100.
Not yet
  • A version deployed at a customer site.
  • A public demo.
  • Public API documentation.
07 — TeamWho builds it
Ziyan
Founder, Lydr Pty Ltd

Ziyan is the developer of Lydr AI and writes the code.

ziyan@lydrai.com
08 — CompanyLegal identity

Lydr Pty Ltd

Lydr Pty Ltd is an Australian software company. Its product is Lydr AI.

Registered legal name
Lydr Pty Ltd
Product
Lydr AI
Incorporated
July 2026
Location
Sydney, New South Wales, Australia
Address
Unit 807/28 Zetland Ave, Zetland NSW 2017, Australia
Ownership
Privately held. No outside funding.
Website
lydrai.com
Contact
Ziyan, Founder, ziyan@lydrai.com
Lydr AI in one paragraph

Lydr AI is a real-time video-intelligence platform for construction sites. It turns the cameras a site already has into a live feed of safety, compliance and throughput events — on-site, in real time. Lydr AI detects and tracks people, equipment, vehicles and zones, and raises structured events: PPE missing, exclusion-zone entry, near-miss, dwell time and throughput counts. Inference runs locally on edge hardware at the site, so video never has to leave the premises, and events are pushed to the tools the team already uses via a simple API. Lydr AI is in development by Lydr Pty Ltd in Sydney, Australia.

09 — QuestionsShort answers

Do we need new cameras?

No. Lydr AI works from the camera and video feeds a site already has.

Does video leave the site?

It never has to. Inference runs locally on edge hardware at the site, and Lydr AI pushes events, not footage, to the team's tools.

What does the team receive?

Structured events: PPE missing, exclusion-zone entry, near-miss, dwell time and throughput counts, pushed to the dashboards, radios and systems the team already uses.

Is Lydr AI a product or a service?

A software product. Lydr AI is a platform with an API that a site deploys.

Can we use Lydr AI today?

Not yet. Lydr AI is in development. No version is deployed at a customer site and there is no public demo yet.

Which NVIDIA technology does Lydr AI use?

In use today: NVIDIA H100 GPUs, to train the vision models. Planned: NVIDIA TensorRT and NVIDIA DeepStream.

10 — ContactZiyan, Founder
ziyan@lydrai.com