AI / Computer Vision Developer – Architectural Drawing PDF Intelligence

We are a commercial millwork engineering company developing an AI-assisted system for interpreting large architectural drawing packages. We are looking for an experienced AI / Computer Vision / Document Intelligence developer to build a focused paid prototype. This is NOT a chatbot project. Our typical input may include a 300–700+ page architectural/interior PDF drawing set plus a scope document identifying a specific millwork or cabinet run. Example: Series 2300 Reception Desk Level 3 Room 312 The prototype needs to identify the architectural information related to that specific scope item, including items such as: Overall floor plan Enlarged plan Interior elevations Sections Details Reflected ceiling plan (RCP) Finish plans and schedules Hardware information Equipment information Other directly referenced drawings A major part of the challenge is following architectural drawing references. For example: Room 312 → A521 → Elevation 4/A621 → Section 7/A742 → Detail 12/A811 The system must recognize and follow relationships like these through a large multi-page PDF drawing package. PHASE 1 PROTOTYPE For this engagement, we are NOT asking for the complete production system. The prototype should: Ingest a large architectural PDF drawing set. Ingest a written scope description identifying a particular cabinet/millwork series. Index the drawing package by sheet number, sheet title and available drawing content. Identify candidate sheets relevant to the selected series. Recognize common architectural references such as 4/A621. Follow those references to additional relevant sheets, sections and details. Rank candidate information by relevance/confidence. Preserve traceability to the original PDF, sheet and page. Produce structured output that can later feed PDF-generation and AutoCAD-automation systems. We will provide completed real-world projects for testing where we already know the correct architectural references. RELEVANT EXPERIENCE We are particularly interested in experience with some combination of: Python Computer Vision OpenCV PDF parsing Vector PDF extraction OCR / text recognition Document AI Multimodal AI models Machine Learning Natural Language Processing Object detection / image analysis Structured data extraction Experience with AutoCAD, Autodesk APIs, DWG/DXF, Revit, architectural drawings, construction documents, engineering drawings or construction technology is a major plus. IMPORTANT TECHNICAL CONSIDERATIONS Architectural PDFs may be native vector PDFs exported from AutoCAD/Revit, raster scans, or a combination of both. We are not looking for someone whose entire proposed solution is simply OCR, RAG, embeddings or a chatbot wrapper. We expect the developer to determine which portions of the problem are best handled using deterministic PDF/vector extraction, OCR, computer vision, rules/geometry, multimodal models and/or LLMs. FUTURE OPPORTUNITY If Phase 1 succeeds, future stages may include: Automatic cropping of individual architectural views Creation of condensed series-specific PDF drawing packages Drawing revision comparison Automated drawing-reference graphs Finish/equipment coordination AutoCAD PDF import Automated scale verification Creation of series-specific DWG mobilization files This could become a substantial ongoing development project for the right person. HIRING PROCESS We expect to select a small number of finalists for a paid technical test using the same real-world sample drawing package. The strongest candidate will then be selected to continue development of the Phase 1 prototype. Please describe the closest project you have personally built involving document AI, computer vision, PDFs, OCR, engineering/construction drawings or similar structured technical documents. Generic proposals that do not address the actual technical problem will not be considered.