This paper reports a full retrofit of an architecture studio in a hot–dry (ASHRAE 3B) context using a human-in-the-loop, multi-objective generative workflow embedded in a participatory design process. We make explicit how our approach differs from prior work by combining (1) direct stakeholder preference injection into the fitness loop, (2) live daylight and CFD simulation in-the-loop and (3) six-month post-occupancy validation.
About 24 stakeholders (students, faculty and facility staff) participated in a four-stage “spiral”: co-framing goals, parametric exploration with NSGA-II (Wallacei X) and EnergyPlus-based daylight/Butterfly-based ventilation models, preference capture and consensus-building. Ten feedback generations guided selection pressure. We combined quantitative performance data with qualitative coding and a six-month post-occupancy evaluation (POE).
The stakeholder-endorsed solution delivered substantial, validated gains: daylight autonomy rose by +29.6 percentage points (38.2→67.8%), natural ventilation approximately doubled (2.8→5.7 ACH), installed cost decreased (∼8%), spatial efficiency improved and user satisfaction more than doubled (2.2→4.6/5). Preference convergence was high (Kendall’s t ≈ 0.86), indicating strong co-ownership of outcomes.
As a single-studio case, external validity is constrained. Transferable elements include the spiral workflow, feedback cadence and low-capex retrofit moves; context-specific parameters (e.g. wind roses and unit prices) may vary. A planned replication across eight identical studios with pre-registered thresholds (e.g. DA = 60% and ACH = 5) will test generalizability.
For public universities in hot-dry regions, the workflow offers a replicable, low-budget path: adopt performance baselines (e.g. DA = 60% and ACH = 5), include HITL workshops as a line item in minor-works procurement and release a shared parametric template with light-touch training for facilities teams.
Participatory co-decision increased perceived agency and satisfaction among students and staff, supporting healthier, more collaborative learning environments.
This is, to our knowledge, the first empirical integration of participatory design with HIL multi-objective generative optimization in a Middle Eastern higher-education retrofit, with live simulation in-the-loop and POE confirmation. The results show that democratic participation and computational optimization can be mutually reinforcing.
