Researchers have developed a framework to convert structural framing plans into editable finite-element model drafts. The system uses an agentic vision-language layer to detect components and draft models from PDFs without requiring task-specific detector training or fine-tuning. The deterministic stage extracts primitives, estimates scale, and recognizes entity classes, while the agentic stage proposes typed corrections based on deterministic candidates. Evaluation showed high recall and precision for detecting columns, beams, walls, braces, and openings in a benchmark of 100 plans. --- Why it matters: This framework matters to engineers working with structural framing plans because it can automate the labor-intensive process of converting these plans into editable model drafts, reducing transcription errors and improving efficiency. It also has implications for researchers developing drawing-understanding systems and language-model agents in structural engineering. Source: https://arxiv.org/abs/2608.17237