•Built an AI-driven cost estimation platform for 3D building models, supporting large model files (~500 MB) across multiple formats.
•Set up self-hosted infrastructure using Docker and DigitalOcean, including S3-compatible MinIO storage and PostgreSQL for model and metadata management.
•Created tRPC and GraphQL routes to fetch model data, integrate with Speckle services, and support seamless model retrieval and processing.
•Implemented material extraction from 3D models and semantic matching against unstructured cost sheets (~5000 rows) using VectorDB and semantic search.
•Improved platform performance, UI flow, and reliability to bring the MVP to production readiness.