SOLARSPLIT
Automated roof analysis for solar — from Swiss LIDAR to iOS.
A freelance AI engineering mission at SOLARSPLIT (Neuchâtel, Switzerland): building the automated roof-analysis engine that works out where solar panels make sense — designed and industrialized end to end, from raw Swiss geospatial data to the iOS app.
Context
Switzerland publishes exceptional open geospatial data: SwissTopo’s swissSURFACE3D LIDAR point clouds and swissBuildings3D models. Buried inside is everything a solar company needs to know about a roof — its usable surfaces, slopes and orientations. The mission: extract those answers automatically, at the scale of the whole country, and put them in the hands of the product.
What I built
A complete AI pipeline, from raw data to production:
Geospatial data & processing
- Massive LIDAR processing — hundreds of thousands of SwissTopo files
- 3D building reconstruction from swissBuildings3D data
- 3D geometry per roof plane: usable surface, slope, orientation
- Automated dataset creation, annotation and preparation for training
AI & computer vision
- Fine-tuning of segmentation models
- Fine-tuning of image-generation models
- Inference optimization: latency, model size, cost
Backend & deployment
- Python / FastAPI AI backends, containerized with Docker
- Serverless GPU deployment
- Continuous tuning of inference cost and performance
Product integration
- AI services wired into the iOS app (Swift / SwiftUI)
- Prototypes hardened into production systems
Across the whole chain
One engineer, the full technical chain: data, models, backend, cloud and mobile. The value isn’t in any single piece — it’s that the pipeline holds together end to end, and that what started as prototypes now runs in production, powering real solar deployment decisions.