Augustin Perraud, Mariano Rodriguez, Peter Starr and Charles Desbaux / Freeda
“It’s profoundly boring and tedious work and it’s the type of thing that you could put 100, 200, 300 hours into it. You’ll always have errors,” Peter Starr says.
He’s telling me about his previous life as an architect, where he had ambitions of grand designs and ended up spending 90% of time on compliance checks and was still never done — something he’s trying to solve with his startup Freeda.
The Paris-based company, which is using AI to spot errors in construction plans, has raised €3.4m from Frst with participation from Brick & Mortar Ventures, combining technology and traditional construction, just as it does with its product.
The round also included angel investors, from early employees at Mistral to business executives from the construction world.
The problem? It is traditionally difficult for AI to read plans. This is because LLMs and general AI cannot interpret architecture plans. It’s not text and it’s not just an image, it’s data and diagrams and text all combined.
If you were to design something using computer-aided design (CAD) and drop the files into ChatGPT, it wouldn’t know what it was looking at. This means checking architecture and design plans is still an incredibly laborious task, and where it can take hundreds of hours of manual verification per project, Freeda detects errors in plans in 48 hours.
“What we’re building is more than AI, it’s computer vision,” Starr says. “It’s building models that understand and that extract information from plans. So a machine on its own understanding that this circle in this context is a table — not a toilet — [and then] we bring AI on top of that to add reasoning and context.”
Catching these errors early could also have a significant environmental impact thanks to less rework and demolition from mistakes in architectural plans. It could also save businesses significant costs given each delayed opening runs the risk of costing more than €100k per month in rent, missed revenue, and loan interest, which is particularly an issue in the hospitality industry.
10,000 plans have been reviewed on Freeda so far since launch and the goal is for one million plans to be reviewed by the end of 2026, including both projected client plans and open-source datasets.
There are also plans to start pre-training the models on new geographies ahead of plans to expand further in 2027. Being global is a part of the startup’s DNA. Freeda has an international founding team — Starr is British American and went to business school in Paris before working at ScorePlay where he met his co-founder and COO Augustin Perraud. CTO Charles Desbaux joined next from Darktrace, followed by Cuban rocket scientist and CSO Mariano Rodriguez.
With six nationalities represented across their nine employees, they have a similarly global client list across Europe, the UK, the US and the Middle East.
Rodriguez described the complexity of architectural plans as “of another order”.
“Plans are one of the hardest data types to tackle from an AI perspective,” Rodriguez wrote on LinkedIn. “Not because they are big, but because they are messy in clever ways. A single sheet mixes vector and raster, old revisions layered under new ones, symbols that look identical until a tiny tail means ‘not a door,’ scales that shift mid set, and details that imply 3D space from 2D ink.”
Whether it’s measuring fire safety routes or ensuring buildings are structurally sound, Freeda is designed to check for all kinds of potential errors.
The company will use the funds raised to grow on both technological and human sides of the business. It plans to develop the first version of a machine that understands architecture plans like a human (which requires more rocket scientists) and also human experts in construction architects in order to recreate the mental workflows that people go through when working on plans.
“Architectural plans are a type of information that seems very intuitive to humans because they’re designed by and for them, but machines struggle with them,” Pierre Entremont, Partner at Frst said. “Changing this is crucial to bringing construction into the AI era.”