← Back to The AFlow
Research

Built on the record.

The AFlow is the build-out of years of peer-reviewed work on computational design, design authorship and spatial topology — not a product with a citation bolted on. The graph, the locks, the provenance trail and the topology engine each trace to published research by the team and its collaborators: Theo Dounas, Elien Vissers-Similon, Yorgos Berdos and Wassim Jabi among them.

4Research threads
15Selected references
2018–26Span
01 / FOUNDATIONS
Four threads

Four lines of research run through it. Each maps to a part of the platform you can use today.

01 · FOUNDATION
The architect as orchestrator
Theo Dounas's doctoral work recast the architect as someone who builds and conducts computational tools — shape grammars generating variations of a complex building — rather than drawing every line by hand.
Grounds: the node graph itself
Dounas · 2019, 2021
02 · SPATIAL MODEL
Topology as the substrate
Wassim Jabi's Topologic and topologicpy model a building as non-manifold topology — a CellComplex of spaces you can query, analyse and run graph measures over. Dounas & Jabi extend it to bridge shape and graph grammars.
Grounds: the Topologic engine & floor plans
Jabi · Aish · Chatzivasileiadi · Dounas · 2018–2025
03 · DESIGN REASONING
Interpreting design, not just images
Elien Vissers-Similon & Theo Dounas study empirically how designers read and steer AI output — classifying AI techniques for early design, and tracing textual, visual and spatial interpretation through diffusion-driven exploration.
Grounds: design memory & OntoCAAD features
Vissers-Similon · Dounas · De Walsche · 2024–2026
04 · GOVERNANCE
Authorship, governed
Yorgos Berdos & Theo Dounas frame design with AI as co-authorship that is “incomplete by design” — and earlier work on data governance sets out how to record who made what and govern what AI may touch.
Grounds: FnF governance & provenance
Berdos · Dounas · Lombardi · McDonald · 2020–2026
02 / METHODOLOGY
Research programme · building now

Spatial analysis with Topologic. One verifiable model, scored by one harness — not another generator chasing a render.

Most AI tools emit pixels, boxes or polylines and try to recover the plan afterwards. We do the opposite: generation writes straight onto one verifiable Topologic model — design, structure, energy and configuration held in a single CellComplex, and scored by one harness.
Brief→ ConstraintGraph→ generator→ CellComplex→ verify→ IFC4x3
One substrate
A model that's also a contract
  • Rooms are cells; shared walls are shared topology — watertight by construction
  • Doors placed as apertures; adjacency & visibility graphs derived, never hand-drawn
  • Fast calculations and variant generation via the Topologic model
Swappable generators
Many methods, one output
  • Grammar, cellular automata and rectangular-dual layout
  • Optimisation (CP-SAT, NSGA-II) and learned samplers (diffusion, GNN, LLM)
  • All write the same substrate; the procedural ones double as the data factory
One verifiable harness
“Valid” isn't “good”
  • Validity, adjacency fidelity, compactness and daylight scored on every plan
  • Graph analysis run on the model — closeness & betweenness centrality, space-syntax integration, isovists and step depth, with a DepthmapX-checked lineage
  • FnF locks gate the result; BuildCheck runs the regs
The governed loop closes it: propose → lift → verify → critique → re-sample, within a fixed iteration bound — and the features you lock are never violated, whoever proposed the change.
03 / BIBLIOGRAPHY
Selected references

The spine. Peer-reviewed papers, book chapters and the library the platform is built on.

  1. Dounas, T. (2019). Shape grammars as an analysis tool of complex buildings [Doctoral dissertation]. Aristotle University of Thessaloniki. doi:10.12681/eadd/45537
  2. Dounas, T. (2021). Decentralised Education. In Five Critical Essays on Architectural Education (A. Williams, series ed.). Machine Books. ISBN 978-0-9572884-9-2.
  3. Jabi, W., Aish, R., Lannon, S., Chatzivasileiadi, A. & Wardhana, N. M. (2018). Topologic: A toolkit for spatial and topological modelling. Proceedings of eCAADe 36, Vol. 2.
  4. Jabi, W. & Chatzivasileiadi, A. (2021). Topologic: Exploring Spatial Reasoning Through Geometry, Topology, and Semantics. In Formal Methods in Architecture (pp. 277–285). Springer. doi:10.1007/978-3-030-57509-0_25
  5. Jabi, W. (2025). topologicpy [Software]. Zenodo. doi:10.5281/zenodo.11555173
  6. Jabi, W., Alymani, A. & Alammar, A. (2024). Enhancing Building Energy Efficiency Estimations Through Graph Machine Learning: A Focus on Heating and Cooling Loads. SSRN. doi:10.2139/ssrn.5034619
  7. Dounas, T. & Jabi, W. (2025). Towards Bridging Shape and Graph Grammars Through Topology. Proceedings of eCAADe 43 (Confluence), Ankara, Vol. 1, pp. 663–672.
  8. Dounas, T., Lombardi, D. & Jabi, W. (2020). Framework for Decentralised Architectural Design: BIM and Blockchain Integration. International Journal of Architectural Computing (Special issue, eCAADe + SIGraDi). doi:10.1177/1478077120963376
  9. Dounas, T., Lombardi, D. & McDonald, H. (2025). Decentralising architectural design through data governance. In Blockchain, Smart Contracts and Distributed Ledger Technologies in the Built Environment. doi:10.1049/PBBE007E_ch8
  10. Berdos, Y. & Dounas, T. (2026). Incomplete by Design: Navigating Co-Authorship Between Architects and AI. In Architecture in the AI Era for Research, Practice, and Pedagogy (SpringerBriefs). Springer. doi:10.1007/978-981-95-0760-3_4
  11. Vissers-Similon, E. & Dounas, T. (2026). The Human Designer in Times of Artificial Intelligence: Diffusion-Driven Architectural Design Explorations. In Architecture in the AI Era… (SpringerBriefs). Springer. doi:10.1007/978-981-95-0760-3_18
  12. Vissers-Similon, E., Dounas, T. & De Walsche, J. (2024). Classification of artificial intelligence techniques for early architectural design stages. International Journal of Architectural Computing. doi:10.1177/14780771241260857
  13. Vissers-Similon, E. & Dounas, T. (2025). Textual and Visual Interpretation in a Text-to-Image Accelerated Architectural Design Process. In Advances in the Integration of Technology and the Built Environment (AAB 2024), LNCE 593. Springer. doi:10.1007/978-981-96-4749-1_18
  14. Vissers-Similon, E. & Dounas, T. (2024). Spatial Interpretation in a Text-to-image Accelerated Architectural Design Process. eCAADe 2024: Data-Driven Intelligence. doi:10.52842/conf.ecaade.2024.2.527
  15. Vissers-Similon, E. & Dounas, T. (2026). Bridging the Gap between Architectural Designers and Developers: Why Empirical Design Research Is Crucial for Computational Design Support. European Conference on Computing in Construction, Corfu, Greece. (DOI forthcoming.)
↳ A-Flow's own substrate — OntoCAAD, the FnF framework and the Design Manifold — is documented in the team's working papers; full citations on request.