What if damage were a prompt — what stories and repair strategies might it generate? This question framed our workshop, which invited participants to work with an AI-supported workflow in which a VLM, accessed through a mobile interface, translated heterogeneous data into spatial models and procedural representations of possible repair interventions. Rather than prescribing a single solution, these proposals made different repair strategies and successive intervention steps explicit, open to comparison and debate.

Working with concrete examples — including doors, windows, chairs, tables, and a damaged pillar from a larger architectural use case — participants examined how different repair decisions respond to feasibility, care, and future use. Through the VLM-backed interface, they engaged in conversation, damage registration, strategy development, and image generation to explore playful and creative repair strategies. The workshop highlighted how AI can support the analysis of existing structures and the development of repair designs, while human expertise remains essential to interpretation and judgment. Repair, in this setting, became something AI could support, initiate, and help explore — without replacing the judgment at its core.

The workshop was developed in collaboration with the ETH Chair of Construction Heritage and Preservation, whose team contributed domain expertise in preservation, material practices, and repair methodologies. Building on their insights, we had in-depth discussions about how repair was understood in the past, how it is approached today, and how it can help extend the future of construction.

Workshop-Team:

Professorship of Digital Fabrication (TUM)
Tizian Rein. M.A.
Begüm Saral, M.A.
Kathrin Dörfler, Prof. Dr.

Chair of Construction Heritage and Preservation (ETH Zürich)
Laurence Crouzet, MArch
Wen-Shan Cui, M.A.
Adrian Pöllinger, Dr.
Silke Langenberg, Prof. Dr.