X:0000PXY:0000PXLIVSYT · 2026

A viaduct is one decision, repeated a thousand times.

AI generative design builds it from the alignment.

Set the megawatt target. Get the building.

Halls, power and cooling, generated as one model.

Then the site has to prove what it built.

Every claim found in the pixels. Or dropped.

Design it. Build it. Prove it was built.

Keep scrolling 01 · LOW WIDE — VIADUCT

One engine, three build types

Rules do the drawing. Engineers do the deciding.

Our AI generative design engine is rule-based, not prompt-based. It reads your alignment, loads, codes and type libraries, then issues the model, the validation, the drawings and the BOQ from one source. Site Intelligence closes the loop on site. Nothing here asks you to model the whole project first.

AI design for metro

Alignment-driven piers, caps, girders, viaducts and station systems. Validated on every change.

livdes.ai/metro

AI design for data centres

Capacity in, hall out: rack grid, electrical and cooling rooms, skids, containment, cost per MW.

livdes.ai/datacentre

Site Intelligence

Progress verified from imagery against the schedule, mapped to BOQ lines, approved by a person.

on LivSYT

Problems

Design is the fast part. Everything around it isn't.

[01]Manual

Every change is redrawn

Move the alignment and the piers, drawings and quantities are all done again by hand.

[02]Blind

No model to check against

Most Indian and Middle Eastern infrastructure has no BIM model, so tools that need one are out.

[03]Disputed

Two versions of progress

Work done on site and work billed are different numbers, argued monthly with photos and memory.

[04]Slow

Weeks lost to repetition

The rule-bound work fills the programme. It is exactly the work software should be doing.

Rule-bound work shouldn't eat your programme.LivSYT takes it. Engineers keep the judgement.

How it works · 01 / 03

It starts with the alignment, not a model.

Chainage, levels, spans, ground conditions. The inputs an engineer already has on day one, and the codes the project is built to.

Scroll to continue

AI generative design builds the structure from the rules.

Piers, caps, girders and deck placed along the alignment, checked for code and clearance, re-checked every time the alignment moves.

Scroll to continue

Drawings and BOQ leave the same model.

No redrawing, no reconciliation. Quantities trace back to the element that produced them, and every revision is versioned.

Scroll to continue
CH 0+000CH 2+240 ALIGNMENT · IMPORTED
7 PIER TYPES · 28 m SPANS · GENERATED
PIER TYPE P3 — GAREV C · ISSUED SCALE 1:50SHEET 04 OF 22 BOQ 04.21 REBARt 04.22 FORMWORK 05.10 CONCRETE 06.02 GIRDERno 06.08 BEARINGSno

[01] · livdes.ai/metro

AI design for metro and rail

A viaduct is one set of decisions repeated a thousand times. AI generative design takes the alignment and the pier library, generates the structure, and re-validates it the moment the alignment moves. Engineers set the rules and sign the drawings. The repetition is what gets automated.

  • Piers, caps, girders, bearings and deck generated along a live alignment
  • Code and clearance checks on every change, propagated to every dependent element
  • Clash intelligence across structural, MEP, station and utility systems
  • Drawings, quantities and BOQ issued from the model, never redrawn

Exchanges data with MiDAS, Oasys, STAAD Pro and Excel

Viaduct generated from alignment + pier type library

[02] · livdes.ai/datacentre

AI design for data centres

Four levels generated from the capacity target: electrical, two data halls, cooling plant

You set
MW target · rack profile · envelope
You get
hall + MEP + BOQ + drawings
Turnaround
days, not weeks

Indicative for a four-level colocation build

Power, cooling, and how many racks fit before something breaks. Set the capacity you're chasing; AI generative design produces the hall, checks it against itself, and prices every version from the model. When the client moves the megawatt target on Friday, the drawing set moves with it.

  • Rack grid, aisles, containment and clearances sized to the capacity target
  • Electrical rooms, cooling skids and distribution generated with the hall
  • Alternatives compared on area, cost and cost per megawatt
  • Versioned BOQ and drawing sets, with rollback

SSO, role-based access, audit trail, on-premise or gov-cloud

[03] · Site Intelligence

Work done on site and work billed are two different numbers.

That gap costs money. Most AI watches a site and guesses. Site Intelligence flips the question: the AI gets each frame together with what the schedule expects in that zone, and must confirm or dispute it with visual evidence. Claims it can't find in the pixels are dropped. An engineer reviews short episodes instead of thousands of frames, and every approval joins a permanent record mapped to BOQ items. No BIM model required.

Drag across the site to verify

Zone view: schedule expectation checked against located evidence

Review queuetoday
Pier P3 — rebar cage14:02–14:04 · 3 frames located · BOQ 04.21confirmed
Span S3 — girder launch09:41–09:46 · 5 frames located · BOQ 06.02confirmed
Pier P7 — shutteringschedule expected · not found in framedisputed
Level 4 west — blockwork11:20–11:23 · partial coverageto engineer
Approach slab — subgrade16:05–16:07 · 4 frames located · BOQ 02.14confirmed

People stay in charge

  • An engineer reviews every reading before it counts
  • Projects earn auto-approval over time; any PM can switch it back
  • The person who reviewed the evidence can't also approve the report — the system enforces it
  • Issued reports can't be edited: identity stamped, evidence sealed
  • Engineers review episodes — "Pier P3, rebar, 14:02 to 14:04" — not sixty frames

For skeptics: this pipeline ran side by side with the old system and took over only when the results agreed.

Who touches it, and for how long

Time in Site Intelligence, per person
PersonWhat they doTime
Project adminSets it up once~30 min, once
Project engineerReviews what the cameras saw~30 min/day
Project managerApproves claims and reportsat milestones
LeadershipReads the digest, tracks slippage~5 min/week

Getting started is the easy part

  • Half a day to one day on site, walking the zones your planner already uses
  • A six-step wizard sets up zones, cameras and coverage; the planner signs off first
  • Start with a phone and a schedule. Add fixed cameras when you're ready

A day with it

Cameras stream. Minutes later, the first detection.

The engineer opens the queue and clears it in 25 minutes.

Progress rolls up against BOQ lines.

The digest email lands with what slipped and why.

The system learns from the day's reviews. Nobody has to do anything.

Site Intelligence is part of LivSYT, the construction project-management platform. LivSYT AI comes with it: an assistant that answers questions like "give me today's DPR" or "what's slipping, and why?" straight from live project data.

Five steps · every claim earns its place

Read

The AI sees the frame and the schedule together and reports what it found. It is answering a question about this zone, not describing a picture.

18 claims read · scroll to continue

Locate

Every claim must be found in the pixels, with a box around the evidence. If it can't be, it's dropped. This is where confident nonsense dies.

13 located · 5 dropped · scroll to continue

Re-check

Independent passes challenge the result. They can relabel it or drop it. Agreement is earned, not assumed.

11 survived · scroll to continue

Record

Only verified claims are written down: zone, activity, confidence, stage. Each one mapped to the BOQ line it belongs to.

11 recorded to BOQ · scroll to continue

Route

Confident results move forward. Everything else goes to a person, as a short episode with its evidence attached.

5 to an engineer · 1 disputed to PM

Who it's for

Built for the people on the project

Bridge engineers

Generate pier and girder families from the alignment instead of detailing each one.

MEP leads

Size power and cooling with the building, not after it.

Design managers

Price alternatives before committing a scheme to the client.

BIM managers

One model issuing drawings, quantities and revisions, versioned with rollback.

Quantity surveyors

Progress claims tied to BOQ lines and backed by dated evidence.

Project managers

Approve verified progress at milestones instead of arbitrating photographs.

Site engineers

Clear the day's review queue in about half an hour.

Leadership

A weekly digest of what slipped, and why, from live project data.

Gurugram metro · ten pier types · same scope both ways

The time saving isn't a demo number.

Design time saved

0%

less design time across ten pier types

By hand 0 h
With AI 0 h
Rework after a design change
~50% lower
Progress claims backed by evidence
every one
BIM model required
none

Questions we get on stand

Straight answers

What is AI design for infrastructure?

Generative design driven by engineering rules rather than prompts. You give the system the alignment, the loads, the codes and the type library; it produces the 3D model, validates it on every change, and issues drawings, reports and BOQ from the same source. Engineers stay the authors — they set the rules and approve the output.

How long does AI generative design take on a metro pier type?

On a Gurugram metro viaduct study covering ten pier types, a 190.5-hour manual workflow was completed in 66.5 hours — a 65% reduction. The saving comes mostly from repeat validation and documentation, not from skipping design decisions.

Can it design a data centre hall?

Yes. Set the target capacity, the rack profile and the site envelope. AI generative design produces the hall grid, electrical and cooling rooms, skids and containment, checks clashes across disciplines, and prices each alternative from the model so you can compare cost per megawatt before committing.

Does Site Intelligence need a BIM model?

No. It works from a schedule and site imagery. Zones use the names your planner already uses — Pier P1, Span S3, Level 4 west wing. Most Indian and Middle Eastern infrastructure projects have no BIM model, which is why the product was built this way.

How does Site Intelligence avoid guessing?

The AI receives each frame together with what the schedule expects in that zone, and must confirm or dispute it with visual evidence. Any claim that cannot be located in the pixels is dropped. Independent passes can relabel or drop a result, and an engineer reviews every reading before it counts.

How long does setup take?

Half a day to one day on site. You walk the site with the planner and record the zones; a six-step wizard configures zones, cameras and coverage, and the planner signs off before anything goes live. Start with a phone and a schedule; add fixed cameras later.

Which engineering tools does it work with?

LivSYT exchanges data with MiDAS, Oasys, STAAD Pro and Excel, so analysis and documentation stay consistent with the generated model. Enterprise deployments support SSO, role-based access, audit trails, on-premise and gov-cloud hosting.

Bring one asset type. We'll design it in front of you.

A pier type, a hall, or one zone of a live site. Forty-five minutes, your data, your codes. If the output doesn't hold up against your own drawings, you've lost an afternoon.

Email the team