Parametric CAD/drafting portal with an AI assistant

Reference site: https://www.rayon.design The goal is to build an AI-driven drafting engine that turns raw data and short prompts into well-structured, publication-ready technical reports automatically. At its core, the system must recognise the user’s inputs, pull from internal style guidelines, and generate coherent sections—introduction, methodology, results, conclusions—without manual intervention. Key functions I expect from the first release: • Natural-language generation that adheres to engineering and scientific writing standards. • Template management so different report layouts can be swapped or updated on the fly. • A feedback loop that lets users accept, reject or edit each paragraph and feeds those edits back into the model to improve future drafts. • RESTful endpoints so the drafting logic can be embedded into an existing web portal; front-end work is minimal, but clean JSON in/out documentation is essential. You are free to leverage large-language-model APIs (e.g. GPT-4, Claude-3) or an open-source LLM fine-tuned with domain data, as long as you clearly state the stack. Python is preferred for orchestration, but I’m open to alternatives if there’s a compelling speed or cost advantage. Please factor in common NLP libraries—spaCy, LangChain, Hugging Face—and modern vector databases for retrieval-augmented generation if necessary. Turnaround is urgent; I need a functional MVP ASAP. A short demo video or live link that shows “data in, report out” will serve as acceptance criteria, accompanied by a concise setup guide so I can reproduce results locally or in the cloud. Once the MVP is signed off, there’s room for follow-on milestones such as multi-language support, advanced citation management, and integration with our internal data lake, so build with extensibility in mind.