Six million points
A laser scan is a cloud of coloured points. It has no walls, no rooms and no dimensions you can build from.
I build systems that turn laser scans, 360° video, photos and drawings into building models and structured data, from the research to the product that runs in production.

This is my scan-to-BIM pipeline running on a real house: six million points in, a measured IFC model out. Every number below comes from that job.






A laser scan is a cloud of coloured points. It has no walls, no rooms and no dimensions you can build from.
Three estimators compete to find which way is up. The winner must leave the floor flatter than it found it.
Everything outside 0.9 to 2.2 metres above the floor fades away, and the floor plan appears from the points themselves.
A vision-language agent draws the walls. Every pass is scored against the cloud and re-prompted with the actual miss.
Geometry decides where it can. Colour decides where it can't, because glass returns points.
Walls rise to their own measured height. The result exports as an IFC building model and a DXF floor plan.
The model, drawn inside the scan from the same camera. Every refinement is scored and rolled back if it makes the result worse.
Want the details: the agent, the scoring, the formats?
Read the scan-to-BIM case studyOne product of my own, research at Aalto University, and long client work in the US and Brazil.
My own product. A laser scan goes in; a DXF floor plan and an IFC building model come out.
Read the case study →
Master's thesis at Aalto University: a photogrammetry pipeline tested in a 24-variant ablation on GPU clusters.
Read the case study →Mezmeriz Inc., Ithaca, New York. Browser tools for dense point clouds and a service that writes native Revit files.
Read the case study →Flopo Ltda., Rio de Janeiro. A Three.js engine with physically based materials for precious metals and gems.
Read the case study →Interactive 3D views of sensor stress and displacement across mining infrastructure.
Multi-user, real-time parametric modeling in the browser for modular construction.
Turns IFC building data into readable JSON and graph cards.
I started in civil engineering and BIM, then spent years writing the software around buildings. Now I build the AI that reads them.
Research in the "Digital Twins of processes for construction of buildings" project, funded by the Research Council of Finland: 360° video to metric point clouds on GPU clusters.
EIT Digital Master School. Entry year at Université Côte d'Azur / Polytech Nice Sophia (2024–2025). Thesis on 360° video to point clouds for BIM.
A SaaS that converts laser scans into DXF floor plans and IFC models, built end to end: point cloud processing, an AI agent, GPU workers, web viewer and billing.
MVP of a real-time 3D jewelry configurator: Three.js engine with physically based materials, REST API and deployment.
Point-cloud editor, BIM modeler and a JSON-to-Revit cloud converter in C#, with AWS infrastructure in Terraform.
3D web work for clients in Spain, the UK, Australia and the US: a retail 3D showroom, a parametric shed customizer, structural health monitoring and remote-sensing analytics.
A cloud platform for on-demand 3D parametric modeling and automated manufacturing of modular construction. Selected for Start-Up Chile (Generation 20).
An online learning platform with protected video streaming and payments. Selected for Start-Up Peru (Generation 4).
BIM coordination and structural detailing for building contractors, with C# scripts on the Revit API to automate modeling.
Focus on BIM, hydraulics, GIS and structural engineering. First place in the faculty's student projects contest (2013).
I trained as a civil engineer in Lima and worked in BIM before I wrote software for a living. Since then I have co-founded two grant-funded startups, spent two and a half years building point-cloud and BIM tools for a US company, and finished an M.Sc. in Data Science at Aalto University with honours.
Today I build computer vision and AI agent systems that work on real data: laser scans, 360° video, photos and drawings. I like owning a problem end to end, from the research question to the GPU worker, the web viewer and the invoice.