AI Material Recommendation System for BIM

What BioBuild says

From BioBuild's Digital Project Manager, in a verified review on Clutch (4.5 of 5; quality 5.0, cost 5.0, willing to refer 5.0):

"Integration of AI reduced the time spent on material selection and design adjustments. We measured this by recording the time taken for these tasks before and after implementing the AI system. The results showed a clear reduction in project timelines."

"We hit some unexpected bumps during the integration, and we are grateful for how quickly the team provided extra support for our on-site team. They weren't in a hurry for fast results, but rather committed to setting us up for long-term success."

How Metwaves Can Help

At Metwaves we build AI systems that sit inside the tools engineering teams already use, rather than beside them. For construction and BIM that means:

• Material recommendation engines like this one, integrated with BIMserver, Revit or your IFC pipeline
• Graph projections of BIM models in Neo4j for fast component and relationship queries (how the projection works)
• Models trained on your own past projects, hosted on your cloud account, retrained as outcomes come in

If you have a BIM workflow and a material, cost or carbon problem inside it, start with a two-week proof of concept against one of your own models. The options are set out on our engagement models page, and the validation approach under PoC to agentic production.

Frequently Asked Questions (FAQ)

Is this a product we can buy, or custom work?

Custom work, delivered and running at one client. The engine, the graph projection and the training pipeline are reusable, so a second build starts from working code, not from zero.

Which BIM tools does it work with?

This build reads IFC through BIMserver, so any tool that exports IFC works: Revit, ArchiCAD, Tekla, Vectorworks. A direct Revit add-in is a short addition if the team wants recommendations inside Revit itself.

How much data does the model need?

Enough past projects to learn the firm's own trade-offs. For BioBuild that was its own completed projects plus supplier and carbon indexes. A firm with fewer projects starts from the public indexes and the model learns the firm's preferences from the pilot's corrections.

Can AI query BIM components in plain language?

It can, and the hard part is not the language model. IFC models are deeply nested, so component questions become slow recursive traversals against a relational store. Projecting the model into a graph makes relationships first-class, which is why this build runs on Neo4j alongside BIMserver. We have not shipped a natural-language BIM query product; if that is what you need, it should start as a proof of concept against your own model.

Can AI prioritise BIM clashes?

Clash detection reliably produces more issues than a team can triage, so the useful work is ordering them: which block work starting on site, which are one modelling error appearing a dozen times, and which will resolve when a discipline re-issues its model. Grouping clashes by root cause needs the same component relationships that make this engine work. This is how we would approach it rather than work we have delivered.

What is session-aware BIM AI?

An assistant that remembers the thread of the current review. Model coordination is a conversation: show me the level four ductwork, then which of those clash with structure, then who owns them. Session-aware means the selected elements, active discipline and filters already applied persist across turns, so a follow-up narrows the previous answer instead of starting again. This is how we would approach it rather than work we have delivered.

How does AI-assisted BIM quality assurance work?

Model QA is rule checking plus judgement, and the two belong apart. Naming conventions, required property sets, missing classifications and orphaned elements are deterministic and belong in code, where results are reproducible. Judgement calls, such as an element classified unlike every comparable element in the project, are where a model earns its place. This is how we would approach it rather than work we have delivered.

How long did this take?

Five months. Work started in August 2024, the pilot was running by October and the engine went live in January 2025.