DataTaxonomy is an automated document intelligence pipeline designed for large-scale architectural and engineering projects. Built for the Fælledby urban development project in Copenhagen, it processes hundreds of project files — PDFs, drawings, models, spreadsheets, and images — and classifies each across seven taxonomy dimensions: domain, scale, lifecycle stage, asset type, information type, governance source, and confidentiality.
The system uses a four-layer pipeline. Layer 1 routes each file to the optimal extraction strategy (text, vision, or metadata). Layer 2 applies an LLM to classify all seven dimensions, enriched by folder-sibling RAG context. Layer 3 scores each file's design decision impact using a geometric mean of four sub-dimensions — constraint hardness, design chain position, phase criticality, and irreplaceability. Layer 4 applies rule-based guards and generates a prioritised review queue.
The result is a live dashboard where project teams can explore their information portfolio, identify knowledge gaps, and focus human review where it matters most — reducing classification effort from weeks to minutes.
