🧠 Systems Thinking in Design: João’s Journey from Architecture to AI, BIM & Computational Design

Our guest at DataDrivenAEC this week is João M. Silva — a leading thinker and builder at the frontier of architecture, BIM, and AI. João’s journey from architectural practice to software engineering is driven by a relentless curiosity and a passion for rethinking workflows through systems thinking and automation. Now working with BuiltCoLAB and the Institute for Advanced Architecture of Catalonia, he’s developing AI-powered tools that support real-time coordination, semantic search, and decision-making in complex AEC environments. Whether it’s building IFC agents or guiding teams through interoperable design systems, João is shaping what the future of data-driven practice could look like.

BuiltCoLAB: https://builtcolab.pt/en/ 

The Elevator Pitch: Why Explore AI in AEC?

DataDrivenAEC: How would you describe your work in one sentence?

João: That’s a tough one! I’d say my work is about discovering how architecture and engineering can benefit from the fast pace of technological development.

DataDrivenAEC: What first inspired you to move from architecture into AI, BIM, and computational design?

João: It happened quite naturally. After a few years of practicing architecture, I was still in love with the discipline but less so with the profession. Especially the repetitive drafting side of it. So I started looking for any opportunity to avoid doing the same thing twice. That began with simple parametric objects in GDL, which taught me about variables. Then I moved on to Rhino scripts parsing incoming BIM models, which taught me about data structures. Eventually, I was developing Grasshopper plugins for colleagues, which pushed me into UI/UX thinking. Step by step, my way of thinking about problems shifted into systems-thinking, and it became hard to go back. That’s when I realized I needed to dive deeper and learn the foundations of computational design, software engineering, and machine learning to move past imposter syndrome.

From R&D to Practice: Bridging the Gap

DataDrivenAEC: In simple terms, what’s the main problem you’re solving with your current AI + BIM projects?

João: I work at BuiltCoLAB, and we do R&D across many areas of AEC, so it’s hard to narrow it down to a single issue. Broadly, I see two main challenges. First, integration: helping teams bridge the gap between traditional, highly structured workflows and algorithms from fields like computer vision, graph theory, 3D graphics or NLP. That’s about making existing processes more powerful and less fragmented. Second, innovation, and that requires deep domain knowledge of how AEC really operates across disciplines, teams, and project phases. Using AI not just to optimize what we already do, but to open new ways of working from the inside out by training, orchestrating and deploying models hand-crafted for our own problems.

On the personal side, my PhD is focused on learning about the design process itself: what makes us iterate, branch, or roll back? How do we evaluate and choose ideas, and why? Answering those questions is fundamental if we want to build systems that augment architects’ design processes. That’s my background, and it’s the foundation for where I want to take this work next.

DataDrivenAEC: Who do you see benefiting most from your work — and how might it change their workflows?

João: My main concern is never straightforward automation. Sure, that makes work faster and cheaper, but not necessarily better. I’m more interested in tools that propose, suggest, or assist. Systems that cooperate with domain experts and help them make better decisions. The people who benefit most are those working at the front line of complex design and coordination: architects exploring options, engineers resolving clashes, or project managers navigating trade-offs. For them, AI can act as a partner that expands their awareness and helps them push beyond “what works” into “what could work better.”

On Data, Agents & the Limits of AI

DataDrivenAEC: What makes BIM and IFC data tricky for AI and agent-based systems to work with?

João: Great question! I wouldn’t say BIM data is inherently tricky for agentic AI. The bigger issue is the lack of high-quality datasets. All the usual AI challenges (unstable frameworks, unreliable outputs, hallucinations, etc.) we inherit as well. But the deeper problem is that most foundational models haven’t been exposed to the specific ontologies and structures of BIM. For example, text-to-SQL benchmarks are optimized for product-type databases, not building data.

That said, I know of promising efforts underway: crowdsourcing the much-needed data, training language models on building ontologies, developing multimodal embeddings that pair geometry with metadata for downstream tasks, and so many others. The pace of research is incredible right now, and it’s genuinely fun to keep up with.

DataDrivenAEC: Can you share a moment when one of your AI agents or MCP experiments surprised you — either in a great way or by doing something unexpected?

João: Definitely! Those are happening more frequently. Recently I’ve been working on a Cursor-like agentic system for BIM project management, where the agent reads and connects across 3D models, timelines, tables, and PDFs. We built an MCP server for IFC model manipulation, and we expected to need heavy in-context learning just to handle schema navigation. But to our surprise, the agent could traverse complex hierarchical relationships with little prior schema understanding, based solely on action / observation patterns and self-correction. These techniques have the potential to transform ordinary queries into semantic search instead of rigid lookups, opening the door to much richer autonomous exploration.

Automation That Actually Matters

DataDrivenAEC: In your experience, where is automation delivering the biggest time savings in BIM workflows?

João: From my experience we see the biggest impact happening in interoperability automation. We have such a fragmented ecosystem of software and vendors that tackling that pain point makes the most sense. APIs are now standard across most services, and with the rise of agentic systems, accessible APIs will become essential, or you risk having your software become obsolete very fast because it can’t integrate vertically with more advanced workflows. Agents are about to become the facilitators of seamless data hoping and feature hunting. We should be no longer tied down to what each software environment offers us. Interesting times ahead.

DataDrivenAEC: How do you balance the need for customization in automation with the need for standardized processes across teams?

João: I think one can’t exist without the other. You always start with standardization, shared processes, a common language, and clear principles. Only then can you build customization on top without breaking the system that supports it.

What’s Next in the AI+BIM Frontier

DataDrivenAEC: What’s one common misconception about using AI in BIM or generative design?

João: That it’s going to replace the creative cognitive load of whoever is in charge. AI can absolutely help you reach the finish line faster and better. But it’s still you who must decide where that line is. Systems that propose “finished” results without your involvement almost always produce poor outcomes that miss the bigger picture.

DataDrivenAEC: For someone starting out, what skills should they focus on to work at the intersection of BIM, computational design, and AI?

João: Start with getting comfortable with programming. I still remember the first few times I could fork GitHub repos and actually run them; it opened a whole new world. From there, just keep building projects and the skills will follow. Always let your curiosity drive you. That’s how you get to the most interesting places.

On a side note, since I see a lot of this in my teaching roles: don’t let LLMs become a replacer for learning. If you put the effort in, they are amazing learning tools. If you’re lazy, they will be your worst enemy.

DataDrivenAEC: What’s next for your research and projects?

João: I’ll follow my own advice and leave that as an open question. I make a point of sharing my discoveries online whenever I can. Keep an eye out — and if anyone reading shares these interests, I’m always open to conversation.

📬 Inspired by João’s approach?
Let us know how you're integrating AI into your design workflows. If you’re working on agent-based tools or tackling design cognition with data, reach out — we’re always looking to feature more real-world innovators.

AEC Tech News This Week

1. Collaborative AECO Practice Model for Healthcare
   A new study discusses a collaborative architecture, engineering, construction, and operations model focused on healthcare facilities in Thailand, highlighting strategies for successful design processes.
   Read more

2. Nemetschek India Establishes AEC Centre of Excellence
   In partnership with VJTI, Nemetschek India is launching an AEC Centre of Excellence to drive innovation and digital advancements in the construction industry.
   Read more

3. AI Empowering Concrete Contractors
   AI-driven tools are transforming scheduling and estimating processes in concrete contracting, integrating data from imagery and predictive analytics.
   Read more

4. Designing AI-Integrated Buildings
   A study highlights the importance of structured datasets for designing buildings with AI integration potential, focusing on data collection in the AEC industry.
   Read more

5. Tesla Outsourcing Advances AEC Innovation
   Tesla Outsourcing Services drives innovation with digital twinning and integrated BIM solutions, addressing data fragmentation in the industry.
   Read more

6. GeoAI Enhancing Construction Workflows
   GeoAI technology improves GIS workflows for AEC teams, allowing for more intelligent integration and data utilization.
   Read more

7. Bentley Systems Launches iTwin Platform APIs
   Bentley has released new iTwin Platform APIs to Cesium Developers, enabling enhanced integration of 3D geospatial data in engineering projects.
   Read more

8. A16z Backing AI Proptech Startup EliseAI
   Raises $250 million to expand its AI tools for real estate and healthcare automation.

   Read more

9. Human-Centric Playground Design
   A focus on human-centered design in playgrounds supports child development with inclusive and participative planning strategies.
   Read more

10. PropTech's Impact on Real Estate
    Innovations in PropTech, such as AI and blockchain, are revolutionizing the real estate sector in Dubai, offering over 97% accuracy in property evaluations.
    Read more

11. Setia Launches AI and PropTech Challenge
    SP Setia Bhd has introduced the Setia AI & PropTech Challenge, advancing smart housing innovations across Malaysia.
    Read more

12. Eco Material Technologies' Low-Carbon Facility
    Eco Material Technologies has opened a new facility in Oregon focused on low-carbon cement manufacturing, contributing to sustainable construction.
    Read more