Titleblock Reader
Finds the title block on an engineering drawing, PDF or DWG, and reads its fields with a vision model that can run fully local.
Every engineering drawing carries a title block: drawing number, revision, author, scale, client. Getting those fields out of a pile of PDFs and DWGs from different suppliers usually means someone typing them over by hand.
Titleblock Reader does it in two steps. A vision model first locates the title block on the sheet and crops it, then reads the label and value pairs from the crop. It talks to any OpenAI-compatible endpoint, so it runs against a local model in LM Studio or Ollama, with an air-gapped bundle for drawings that may not leave the building, or against a hosted model when that is acceptable.