Before puniq, founder Waddah Zekri worked on computer vision and machine learning at elevait GmbH in Germany, building the systems described here. We share it as a case study because it shows the discipline behind puniq: take messy, vendor-bound, undocumented data and turn it into something readable, searchable and durable. The domain was construction. The principle is the same one we apply to buildings today.
Large construction and infrastructure projects, from railways to real estate, generate thousands of site plans. Most of them live as scanned paper, PDFs and CAD exports with no consistent coordinate system. A plan that says "north wing, axis 7" means nothing to a map. When an engineer needs to find every drawing that touches a specific location, the answer is a manual hunt through archives that can take hours per query.
- Plans were stored as flat images with no spatial meaning. You could open a drawing, but you could not ask "what is here?".
- Different contractors used different scales, legends and rotations, so no two archives matched.
- Knowledge lived in the heads of a few senior staff. When they were unavailable, retrieval stalled.
- Auditing and handover required someone to physically reconcile drawings against the real site.
The archive was full of answers. None of them could be found by asking a location.
