The Challenge
What Mabat3D Was Facing
Mabat3D renders detailed 3D city models for urban planners and infrastructure teams, working with datasets that run into hundreds of gigabytes per city. Delivering interactive 3D visualisation in a browser at acceptable frame rates required solving both a data engineering problem — efficiently tiling and serving massive 3D datasets — and a front-end rendering problem, without asking users to install any software.
The Solution
What We Built
We designed the data pipeline to process raw LiDAR and photogrammetry inputs into 3D Tiles (the OGC standard) using a serverless processing pipeline on AWS Batch. Processed tile sets were stored in S3 and served via CloudFront with range-request support, enabling the browser client to stream only the tiles visible in the current viewport. The WebGL rendering engine used frustum culling and level-of-detail switching to maintain 60fps on typical enterprise hardware. Infrastructure for large city ingestion jobs was auto-provisioned via Spot instance fleets, reducing processing cost by over 60% versus on-demand.
Results