Post by Pouria Ahmadi

Archi-Gen V2 | Rethinking Computational Generative Design University Project from 2021 One year after developing the first version of Archi-Gen, Reza Taghavifard and I revisited the project as part of our computational design research at the University of Tehran, under the supervision of Dr. Ali Andaji Garmaroodi, with a simple question: What if we stopped generating geometry first ❓ The initial prototype was built entirely in Grasshopper. It generated hundreds of floor plan alternatives using geometric operations and then evaluated each solution to identify design issues such as inaccessible rooms, missing windows, or inefficient layouts. While the system worked, the workflow had two major limitations: 1️⃣ Grasshopper's computational process became increasingly slow as the number of generated alternatives grew. 2️⃣ The algorithm spent significant time creating floor plans that were later discarded because they violated architectural rules. ✨ For the second version, We completely redesigned the computational workflow. - Instead of relying on geometric operations from the beginning, the system first builds an abstract spatial model using mathematical matrices and Python. - Through an interactive questionnaire, it captures the user's requirements and translates them into spatial relationships before any geometry is created. - Inspired by Ant Colony Optimization, each functional space (living room, kitchen, bedroom, etc.) behaves as an independent agent aware of its neighboring spaces. This allows the system to organize spatial relationships before producing a floor plan. Once the optimal solution is found, the data is automatically exported as a CSV file and imported into Grasshopper, where the final architectural model is generated with walls, doors, furniture, and a simple 3D visualization. ✨ The result was a dramatic improvement: - Generation time decreased from approximately 14 minutes to around 8 seconds. - The system now supports sites of any dimensions, instead of being limited to a fixed 10 × 20 meter plot. - Users still receive multiple design alternatives, but with a far more efficient computational workflow. Looking back, this project taught me an important lesson: 📍 Sometimes the biggest performance improvement doesn't come from optimizing the code; it comes from rethinking the algorithm itself. [Link to Archi-Gen V1: https://lnkd.in/es7ivRNz ] #ComputationalDesign #GenerativeDesign #Python #Grasshopper3D #Architecture #DesignTechnology #AEC #Algorithm #ParametricDesign

Video Content