Recently, an interesting job vacancy caught my eye on LinkedIn: ADYD Group is looking for a “SENIOR REVIT ARCHITECT” for their team in Spain, requiring 15-25 years of experience and proficiency in English/German. And you know what I thought when I saw this list of requirements? I thought about how desperately we in the IT industry cling to the idea of an indispensable human expert, even when technology is already poised to offer an alternative.
**Briefly about the company:** ADYD Group is a reputable engineering firm from Spain. They specialize in design, outsourcing, and training. Judging by their projects and approach, this is not a startup but a mature business that values quality, stability, and, as they themselves write, “the best talent.”
**What pain point are they trying to address?**
Let’s face it. What is an architect with 25 years of experience, who must define structural details, make technical decisions, and be fluent in Revit?
This isn’t just an employee. This is a “walking database,” a living archive of thousands of successful and failed solutions he’s witnessed on construction sites in Germany, Spain, or elsewhere. He remembers how a particular facade joint behaves after 10 years of operation, which material is best not to combine with another in a humid climate, and how to explain to a contractor on site that their “brilliant” simplification idea will lead to disaster.
*— Mikhalych, remember how we struggled with the curtain wall-to-roof junction on that project in Düsseldorf? — one manager asks another.*
*— How could I forget! — Mikhalych replies. — The design called for a standard joint, but in reality, due to supplier tolerances, we had to rework everything on site. If it weren’t for Hans, who’s been through it all, that building would still be standing with a hole in its roof.*
This “Hans” is exactly who ADYD Group is looking for. They are looking for a person whose mind holds unique, unstructured experience. The company’s pain point is risk. The risk of making an incorrect technical decision that will cost millions of euros. The risk that a young BIM modeler, perfectly proficient in Revit, will draw a beautiful but unviable joint. They are trying to buy insurance in the form of one person’s 25 years of experience.
The problem is that such “Hanses” are few and far between in the market. They are expensive, temperamental, and their knowledge will retire with them.
**But what if there’s another way?**
Of course. We can stop hunting for unicorns and start building a system. Instead of searching for a unique knowledge holder, we can create a digital system that will possess, multiply, and share this knowledge with the entire team.
Imagine not a “Hans,” but an “AI Architect Assistant.”
**Approaches, Tools, and Implementation Steps**
This is not science fiction, but a very concrete roadmap for a modern engineering business.
**Step 1: Creating a Knowledge Base.**
Our “Hans’s” brain is his experience. The AI’s brain is the data it’s trained on. First, we need to digitize and structure all the company’s accumulated experience:
* **Data Collection:** Upload all existing projects into the system: drawings (DWG, PDF), BIM models (RVT), technical specifications, estimates, construction site reports, memos about problems and their solutions.
* **Tools:** RAG (Retrieval-Augmented Generation) technology is ideal here. We create a vector database of all company documentation. When an engineer asks a question (“suggest a composite panel facade attachment detail for wind zone III in Germany”), a large language model (LLM), like GPT-4 or Llama, doesn’t invent an answer but finds relevant examples from *your own* successful projects, analyzes them, and proposes a solution based on *your* real-world experience.
**Step 2: Generative Design and Automation.**
Our AI assistant doesn’t just answer questions; it actively helps with the work.
* **Approach:** Instead of manually drawing every detail in Revit, the engineer describes the task and constraints: “I need a parapet-to-flat-roof junction detail. Materials: concrete, specific insulation, specific membrane. Consider DIN 18531 standard requirements.”
* **Tools:** Generative design scripts (e.g., using Dynamo for Revit or Rhino/Grasshopper, enhanced with AI) can generate dozens of detail variations, each analyzed for compliance with norms, heat loss, material costs, and installation complexity. The engineer is no longer a drafter but a curator who selects the best solution from those proposed by the system.
* **Reducing Distrust:** Initially, the system operates in “advisor” mode. It doesn’t make decisions but offers options with detailed justifications: “Option A is the cheapest but requires complex installation. Option B is 15% more expensive but meets energy efficiency class A+ and has been used repeatedly in projects X and Y.” The decision remains with the human, but it is now data-backed.
**Step 3: Integration and Learning.**
The system must be dynamic.
* **Feedback:** After each project, engineers add information to the database about how the implemented details performed on site. “AI suggested detail #34. During installation, it turned out the contractor lacked standard fastener length. Correction Z was made.” The system learns from this data and will propose an improved version next time.
**How to Validate AI Output?**
The most important question is: how do we trust a machine in such a critical field? The answer is simple: the same way we trust a new employee, only the verification process is much more transparent.
1. **Traceability:** For every proposed solution, the AI must provide a source reference. “This detail is based on the ‘Berlin Business Center, 2019, sheet AR-205’ project and complies with clause 5.2 of standard EN 13830.” No magic, just facts.
2. **Digital Simulation:** Before sending a drawing to the construction site, the AI-generated detail can be automatically sent for verification in analysis software (e.g., for Finite Element Analysis — FEA). The system can calculate its load-bearing capacity, thermal conductivity, and resistance to wind loads. This is something no architect does “in their head” for each of thousands of details.
3. **Human-in-the-Loop:** At the final stage, the AI’s output is always reviewed by the lead project architect (LPA). But their role changes. They no longer spend time on routine drafting and solution finding but focus on high-level validation and strategic decision-making, with all the AI’s analytics at their fingertips.
Instead of looking for one person with 25 years of experience, ADYD Group could hire two talented engineers with 5 years of experience and invest the saved salary fund into creating a system that, within a couple of years, would possess the cumulative experience of the entire company.
This doesn’t negate the value of human experience. It allows it to be digitized, scaled, and transformed from a vulnerable asset in one employee’s head into an indestructible capital for the entire company.
Источник: https://www.linkedin.com/jobs/view/4402863916/
