Newforma Acquires Lumi AI.

Ask anyone in your firm how a project is actually doing and you’ll get a feeling, not an answer. The answer exists. It’s sitting in your project records right now.

Getting it out takes someone with time, a reporting tool, and a question specific enough to build a report around. So most of the time nobody asks.

That’s the problem we just bought a company to fix. Newforma has acquired Lumi AI, a team in Toronto that spent three years building an engine you can ask questions of in plain language. No SQL. No report builder. Not a chatbot bolted onto a database either, but an engine that understands what your data means before it answers.

Think about the questions you’d actually want answered on a live project:

• How is this project tracking right now, and what moved since last week?

• Which consultant is the bottleneck on RFI turnaround right now?

• Which of my projects has an issue pattern that looks like the one that went sideways last year?

• Why is this submittal cycle running three days longer than our firm’s baseline?

None of those are hard questions. They’re just questions nobody has time to build a report for. And a report answers one question, once, for whoever asked it.

Pointing a language model at a database is easy. Getting a right answer back is not. On Spider 2.0, a benchmark of 632 real enterprise data questions, the best system tested scored 21.3 percent. The same approach scores above 90 percent on simplified test databases. Real schemas are where these systems come apart.

In construction, a wrong number ends up in a decision that carries liability, so a confident guess is worse than no answer.

Lumi’s engine does three things that matter here. It builds a semantic layer, so it works from meaning rather than column names. It asks you a clarifying question when your question is ambiguous instead of picking an interpretation and running with it. And it tells you when the underlying data has a problem, which is something your project records absolutely will have.

That last one is less exciting than it sounds and more important than it sounds. An engine that tells you the change order values are missing for March is worth more than one that quietly averages around the gap.

Right now, every software company serving this industry is claiming its AI is better than the next one’s. Almost nobody is saying what better means. So here’s our standard, and you should hold us to it.

• Explainable. The answer shows you how it got there. If you can’t follow the logic, it isn’t an answer, it’s a guess with formatting.

• Verifiable. You can trace it back to the source records. Click through to the RFIs it counted.

• Accurate enough to act on. Not accurate enough to be interesting. Accurate enough to use when a decision carries professional liability.

Three out of three, or it doesn’t ship to you. That’s a harder promise than claiming superiority, which is exactly why it’s worth making.

An engine like this learns a customer’s data model quickly, but it needs structure underneath to be trustworthy. That’s our side of the deal. The way Newforma Konekt and Newforma Project Center organize project information carries twenty years of AECO workflow thinking in it. An RFI isn’t a generic ticket. A submittal cycle has a shape. A coordination issue means something specific.

Their engine plus that structure is the thing neither of us had alone.

There’s a word getting used loosely in this industry right now, and it’s context. Every vendor claims to give you it. Most of them mean context from their own products. Your project doesn’t live there. It lives across your design tools, your document system, your email and the field, which means true context has to come from all of it. One view into every source of truth you already have, not one more source of truth asking you to move in.

I should be clear: we bought two things, and both of them move the date. The engine is working IP, and bringing it in pulls insight delivery in Newforma Konekt forward by 18 to 24 months… The IP gets customers and industry years ahead of where our roadmap had this. The people are how it keeps going from there. Tudor Boiangiu joins us as VP of AI Engineering with his team of engineers, and they’ll work directly on the agentic framework our own group has been building.

The direction is self-generating insight. Not dashboards you configure and then forget to open, but a product that watches your projects and surfaces what matters without being asked. In practice, that means Lumi’s engine extending Vojo, the assistant we showed at Newforma World in May, inside Newforma Konekt’s smart project dashboards.

If you want to follow where this goes, talk to your account team. They hear about each step before we write about it.