Architect and AI/ML researcher at NAME Architecture, Paris. Master in AI for Architecture and Business Innovation, IAAC Barcelona.
I design and build AI/ML workflows that solve problems in the AEC industry; from crowd safety and building performance to environmental analysis and robotic fabrication. My work sits at the intersection of spatial thinking and machine learning: I bring an architect's understanding of buildings, cities and data to every model I train.
Currently building NAMELab ; an AI design intelligence platform for NAME Architecture and an automation tool for Encore that generates 3D design options for converting vacant office buildings into residential and mixed-use without demolition.
Internal tools built for practice use (code is private.)
NAMELab is an AI platform developed for NAME Architecture to make accumulated studio knowledge accessible and useful during design.
| Tool | What it does | Stack |
|---|---|---|
| Studio Image Library | CLIP semantic search across 10,000+ architectural images; search by spatial quality, materiality, mood or light. Each image tagged by GPT-4o Vision with 15 architectural attributes. Deployed internally. | Python · FastAPI · CLIP · GPT-4o · UMAP · Docker |
| Design Intelligence | Extracts design reasoning from audio recordings, site notes and project documents into a structured knowledge base. Analyses site context and generates calibrated briefs for new projects. | Python · FastAPI · LLM · RAG · Docker |
| Internal LLM Platform | Secure multi-stack AI system with separate knowledge bases for material research, project history, RFP writing, journal drafting and prompt generation. | Python · RAG · Ollama · Msty.studio |
| ComfyUI Workflows | Image generation and enhancement pipelines calibrated to the studio's visual language. | ComfyUI · Stable Diffusion · Checkpoints · loRAs |
| Encore Automation | Automates the generation of 3D design options for Encore; an adaptive reuse system (NAME Architecture consortium) that converts vacant office buildings into residential and mixed-use without demolition. | Python · JavaScript · Grasshopper · FastAPI |
| Project | What it does | Stack |
|---|---|---|
| Disperse | Crowd density estimation from satellite imagery for event safety planning | Python · FastAPI · React · Keras · Mapbox GL JS |
| RetroGrade | Building facade retrofit scoring; semantic segmentation, climate data, energy regression, PDF output | Python · SegFormer · FastAPI · scikit-learn |
| Surround | PM2.5 air quality prediction from urban land use and ERA5 weather across Krakow; R²=0.85, MAE=6.48 µg/m³ | Python · scikit-learn · Google Earth Engine · ERA5 |
| UrbanShade | Draw any area in Europe, run a UTCI thermal comfort simulation, test cooling interventions | JavaScript · Grasshopper |
| upcycling-timber | CV pipeline for detecting salvageable timber and exporting robot pick-up coordinates | Python · OpenCV · Roboflow · ArUco markers |
| Clay-Create | AI-adaptive clay fabrication; depth camera scans tiles and regenerates designs for visual continuity | Python · Computer Vision · Robotic Fabrication |
ML / Deep Learning — TensorFlow · Keras · scikit-learn · Hugging Face (SegFormer) · Random Forest · Regression
Computer Vision — OpenCV · Roboflow · ArUco markers · Semantic segmentation · Depth cameras
Backend — Python · FastAPI · REST APIs · Uvicorn
Frontend — React · Vite · JavaScript · Mapbox GL JS · React Konva
Geospatial / Remote Sensing — Google Earth Engine · GeoPandas · Sentinel-2 · ERA5-Land · UTCI simulation
BIM / IFC — IFC parsing · Agentic IFC tools · BIM data pipelines
Fabrication — Robotic arm integration · Grasshopper · Rhinoceros
Data — pandas · NumPy · Jupyter Notebooks · Matplotlib
LinkedIn · Paris / Barcelona
