Architecture and construction
Train clash, hazard and defect detection models
Architecture firms, construction-tech companies and equipment makers: train on drawings, BIM models, site cameras and scans.
What it's worth
Cut the cost of rework and site injuries
- of project cost goes to rework on average, within a 2–20% range
- About 5%
- saved on a USD 1 billion build when rework falls from 5% to 4%
- USD 10 million
- estimated cost of one fatal work injury, across all US industries
- USD 1.54 million
Sources: Construction Industry Institute, via Curtin University review; National Safety Council Injury Facts, 2024 estimate; the second figure is arithmetic on the first.
Use cases
Design, site and machine models
| Use case | Raw data | Rare cases to find | Adapt per | Release test |
|---|---|---|---|---|
| Design checking | Drawings, BIM models, specs | Clashes between disciplines, code breaches | Building code, client standard | Issues found against a checked set |
| Site safety | Site and crane cameras | People in swing radius, no harness at height | Site, phase | Missed hazards on replayed footage |
| Progress tracking | Drone flights, 360° walkthroughs, laser scans | Out-of-sequence work, deviations from BIM model | Project, building type | Progress against surveyed quantities |
| Quality inspection | Photos and scans of concrete, steel and welds | Cracks, exposed rebar, bad welds | Material, contractor | Missed defects on an inspector-checked set |
| Machinery | Machine cameras, lidar, telemetry | People in blind spots, unstable ground | Machine type, site | Collisions in simulation |
Design checking
- Raw data
- Drawings, BIM models, specs
- Rare cases to find
- Clashes between disciplines, code breaches
- Adapt per
- Building code, client standard
- Release test
- Issues found against a checked set
Site safety
- Raw data
- Site and crane cameras
- Rare cases to find
- People in swing radius, no harness at height
- Adapt per
- Site, phase
- Release test
- Missed hazards on replayed footage
Progress tracking
- Raw data
- Drone flights, 360° walkthroughs, laser scans
- Rare cases to find
- Out-of-sequence work, deviations from BIM model
- Adapt per
- Project, building type
- Release test
- Progress against surveyed quantities
Quality inspection
- Raw data
- Photos and scans of concrete, steel and welds
- Rare cases to find
- Cracks, exposed rebar, bad welds
- Adapt per
- Material, contractor
- Release test
- Missed defects on an inspector-checked set
Machinery
- Raw data
- Machine cameras, lidar, telemetry
- Rare cases to find
- People in blind spots, unstable ground
- Adapt per
- Machine type, site
- Release test
- Collisions in simulation
How it works
Drawings and site footage in. Tested releases out.
Melt dataPrepare the data · 01–03
01Ingest
Load drawings, BIM models, site cameras, drone flights and laser scans; drop corrupt files, catalogue by project, phase and date, and anonymise faces and plates.
02Mine
Find clashes, near-misses, unsafe acts and defects in drawings and weeks of site footage, and drop near-identical files.
03Label
Auto-label clashes, people, machines, protective gear and defects with large labelling models that run inside the platform; your engineers correct a sample.
Forge intelligenceTrain and test the model · 04–06
04Train
Retrain your shared base model on the full, versioned dataset when data from a new site lands; every release traces to its exact data.
05Adapt
Fine-tune the base model for each building code, site, phase and machine fleet, and tune it again as the project changes week by week.
06Validate
Optimise each release candidate for site cameras and drones, then replay past clashes, incidents and inspector-checked defects against the last release.
Drawing to site
One BIM model for design and site
Design checks read the BIM model; progress and safety models compare drone flights and site cameras against it.
Quality
Regressions stop at the gate
Every release candidate runs the last release's test sets; a drop beyond your margin blocks it until you sign off.
Your team sets the metric and margin for each test set, such as missed clashes, missed hazards, false alarms per hour and missed defects, each with its 95% confidence interval.
Architecture and construction
release v13 → release candidate v14 · lower is better
| Test set | Cases | Last release | Release candidate | Margin | Result |
|---|---|---|---|---|---|
| Missed clashes, checked design set | 420 issues | 18 | 14 | +2 | Pass |
| Missed hazards, replayed footage | 300 clips | 9 | 14 | +2 | Fail +5 |
| False alarms per hour | 50 h | 0.6 | 0.4 | +0.2 | Pass |
| Missed defects, inspector-checked set | 900 photos | 27 | 22 | +3 | Pass |
Missed clashes, checked design set
- Cases
- 420 issues
- Last release
- 18
- Release candidate
- 14
- Margin
- +2
- Result
- Pass
Missed hazards, replayed footage
- Cases
- 300 clips
- Last release
- 9
- Release candidate
- 14
- Margin
- +2
- Result
- Fail +5
False alarms per hour
- Cases
- 50 h
- Last release
- 0.6
- Release candidate
- 0.4
- Margin
- +0.2
- Result
- Pass
Missed defects, inspector-checked set
- Cases
- 900 photos
- Last release
- 27
- Release candidate
- 22
- Margin
- +3
- Result
- Pass
Release candidate blocked: 1 test set beyond its margin
Interface preview · sample data
Your data
Your project data and weights stay yours
- Ownership
- Your data, labels and trained weights stay yours. Bytemelt has no access to them: your contract says so, and the platform is built so only accounts your team controls can open them.
- Anonymisation
- Your team anonymises faces and plates at ingest with the platform's tools, using rules you set, or before the files leave your own systems.
- Location
- Your data is stored and processed only at the partner data centre your contract names, in the Middle East and outside the EU. The contract names the site and country before any data moves, and data leaves the site only when your team moves it. Personal data from the EU moves only under transfer safeguards such as Standard Contractual Clauses, and we give your data-protection officer the site details a transfer impact assessment needs.
Questions design and construction teams ask
We're a contractor. Is Bytemelt for us?
If you want models of your own, trained on your site data, yes: the platform labels, trains and tests them automatically, with compute power included. Bytemelt sells the platform, not model training to order.
Do design checks and site models share data?
Yes. Both train on one copy of your project data: drawings, BIM models, drone flights and site footage.
Size compute for your project data
In one hour, the founder maps your drawings, site footage and models to the six stages and sizes the compute.
contact@bytemelt.ai