On Monday 25 August 2026 we open our doors again — this time to a screen. Pablo Levine Mardones presents his Building System Specialist final project live, and with it he becomes the 39th certified BSS to come through this house. We say that number plainly, not as a headline: thirty-nine specialists across every edition we have run, each with a capstone we could point to. Pablo’s is the one you can watch being defended, in real time, on our stream.

The project itself we covered earlier this week from the outside: Hubert Klumpner’s Vertical Gym typology, developed through Urban-Think Tank in Caracas and now migrating to two towers under construction in Medellín, with a third in early design. Pablo — seven years inside the Chair of Architecture and Urban Design at ETH’s D-ARCH — took four programmes (sport, culture, services, circulation) and made each one a set of parameters instead of a fixed plan. He then handed the search to David Rutten’s Galapagos, the genetic solver that ships inside Grasshopper, and connected the surviving configurations back into a BIM environment for the Swiss–Colombian coordination team.

That is the announcement. The graduation is a different thing, and it is the thing worth an hour of your morning.

←TODAY: A genetic solver proposes hundreds of tower layouts; a human still signs the one that gets built. →3012: The office that survives is the one whose architects can say why they overruled the machine. Fulcrum: Optimisation is only trustworthy when someone in the room can explain where it was wrong.

What to actually watch for

A slick project video shows you the winner. A graduation defence shows you the argument. When you tune in, ignore the render and watch for three moments. First, the point where a committed site constraint becomes a parameter — a municipally owned plot, a setback, a floor-to-floor that could not move, translated from a rule into a number Galapagos can read. Second, the moment Pablo overrules the solver and can say why: the fittest option lost to a human judgement about how a reggaeton-anchored cultural floor actually wants to sit against a sports hall. Third, and rarest in any presentation — what failed. The definition that gave garbage, the fitness function that optimised the wrong thing. A BSS who can name the dead ends learned more than one who only shows the trophy.

This is the pedagogy we describe on the BSS Title page and in our past student stories: build the logic system, don’t just draw the walls; and above all, understand the tool well enough to argue with it. Automating the search is the easy half. Reading its output like a professional is the title.

Atelier: For a Büro adopting generative search this year, the risk is not that the solver is wrong — it is that no one keeps the record of why a proposal was accepted or killed. This week, add one column to your options log: next to every Galapagos or Grasshopper variant your team advances, write the single human reason it beat the fitter one. That one habit turns an opaque optimisation into a decision your client, your Bauleitung, and your future self can audit.

Hack: Read the moment your solver stops earning its keep. A genetic run improves fast, then crawls — and that plateau is exactly where a Galapagos-style search should hand the pen back to the architect. Log the best fitness per generation and watch the gain per step; when it drops under a threshold you set, stop trusting the machine to find meaning it no longer has. This tiny convergence check is the difference between running 400 generations out of faith and stopping at the one where judgement takes over.

history = [8.4, 6.1, 5.2, 4.9, 4.85, 4.84]  # best fitness per generation, lower = better
plateau = next(i for i in range(1, len(history)) if history[i-1] - history[i] < 0.05)
print("gain stalls at generation", plateau, "value", history[plateau])
print("hand the pen back to the architect here")  # -> generation 4

You can run the same logic against the tender’s quantities, a daylight sweep, or any parametric loop that connects Rhino/Grasshopper to Archicad automation — the point is to make the stopping decision yours, on the record, not the solver’s by default.

Then the second reason to watch: we open Edition XIV on 1 September 2026 — five months, six hours a week, ending in exactly the kind of defence Pablo is about to give. If you have wondered what the finish line looks like before signing up for the path, watch someone cross it first. Bring one question you would ask Pablo about a constraint you cannot solve in your own office — and answer it on Monday.

Sources & Further Reading

• Primary: PAZ Academy live stream — Pablo Levine Mardones, BSS graduation
• PAZ context: Building System Specialist Title — programme structure