AI Developer (Claude/Gemini) for Automated PDF Extraction from HVAC Blueprints

Project Overview: I am looking for an AI automation and data extraction expert to build a reliable pipeline that extracts complex sheet metal HVAC ductwork sequences and specifications from architectural blueprints. The goal is to automatically process these blueprints and output the data into a clean, structured master duct-schedule in Google Sheets/Excel so I can clearly understand the system layout and fabrication requirements. The Problem: I currently use Claude and Gemini for this, but I consistently hit roadblocks. The files are large and visually complex. The AI frequently "forgets" the file, drops context, or refuses to process the PDF due to limitations in the standard chat interface. I need a robust API or scripted solution to bypass these web-interface limits. File Formats: We are primarily dealing with large PDF blueprints. However, I can often provide the original AutoCAD (.dwg/.dxf) files. If you know how to extract the raw metadata/layers from AutoCAD before passing it to the LLM, that is a massive bonus. What You Will Do: Build a script/pipeline (likely Python) to parse complex HVAC blueprints (PDF or AutoCAD). Integrate the Claude or Gemini API to accurately interpret the extracted data (identifying duct sizes, lengths, tags, and sequences). Engineer the perfect prompt/system instructions to handle construction terminology without hallucinating. Format the final output so it easily ports into a structured spreadsheet. Ideal Candidate: Expertise in Python, API integration (Anthropic/Google), and handling large context windows. Strong experience with OCR (for PDFs) and programmatic CAD extraction (for .dwg files). Experience dealing with complex tables, schematics, or blueprints. Clear communicator who can build this so it is easy for a non-programmer to run on their own machine. Budget Structure: I am setting a fixed price for the initial successful setup, testing, and delivery of the working script. Once the pipeline is built, I would like to transition to an hourly contract for ongoing support, edge-case troubleshooting, and fine-tuning as I process different types of blueprints.