Original Summary

I work in construction, not software development. But Claude’s basic paid plan has already helped me build tools that automate parts of my actual job. Now I want to tackle the part that takes the most effort. We prefabricate MEP racks: frames carrying pipes, ducts, and cable trays that are assembled off-site and installed in buildings as modules. Before we can manufacture anything, we have to read the architectural, structural, and building services drawings, work out where these modules could fit, and determine what goes inside them. I’ve just received a new project. An entire building, delivered in 2D AutoCAD drawings. No 3D models for any discipline. Another project to understand from scratch. What I’ve already built Using Claude and Python, I’ve created tools inside Revit that automate repetitive modelling tasks and fabrication documentation. These are tools I actually use at work. I’ve also experimented with getting AI to extract information from AutoCAD drawings and generate a 3D model of the building’s structure. That last part is promising, but the reliability is still questionable. I’ve been using an OpenAI model as a second reviewer, which helps, but agreement between models isn’t proof of accuracy. So I’m not starting from zero, and I’m not claiming I’ve solved it. What I want to build next A workflow where an AI agent can: Inspect the drawings across disciplines. Extract the information relevant to our prefabricated modules. Identify constraints, contradictions, and missing information. Produce an initial coordinated 3D layout for review. Reuse the corrections and decisions we make along the way. The engineering designs already exist. I’m not asking AI to calculate how much cooling the building needs or size its plumbing from scratch. I want it to interpret what has already been designed and help us turn it into something we can manufacture. That interpretation and initial layout are the hardest parts of my workflow. Once they’re resolved, the automation I’ve already built becomes much more useful. Why I’m paying for it myself My employer declined to fund Claude Max 20x. I’m going to pay for it because the cheaper plan has already produced enough useful results to make me want to test the next step. I don’t think a more expensive subscription automatically fixes the problem. Better access to models is only part of it. The real challenge is giving the agent the right tools, preserving project knowledge, and checking whether its interpretation matches the drawings. I’ll be testing this on a new project and building around the problems that actually appear. Is anyone here building AI tools for similarly messy, specialised work? Not necessarily construction. Manufacturing, engineering, logistics, technical documentation — anything where the input is more complicated than a clean database and a text prompt. I’m particularly interested in: Moving from useful one-off scripts to a dependable workflow. Giving agents access to technical files t

  • 情报分类：工作与职业机会
  • 分类依据：内容涉及招聘、求职或职业发展
  • 信息来源：Reddit · SideProject
  • 发布时间：2026/10/15 00:10:28