I need a working Deepseek LLM setup that can automatically read incoming documents, extract the text as well as embedded numbers, and return clean, structured data we can feed straight into our internal workflow. The goal is a business-ready solution, not a research prototype, so reliability and clear hand-off documentation are key. Scope • Fine-tune or expertly configure Deepseek so it understands common document layouts (PDF, DOCX, scans) and handles mixed text-numerical content. • Build an inference pipeline in Python—HuggingFace, LangChain, or similar—that receives a file, runs OCR if needed, calls the model, and outputs JSON. • Implement basic post-processing: field validation, simple calculations, and error handling. • Package everything in Docker with a concise README so our DevOps team can deploy to AWS and deploy it to our contabo deepseek private environment. • Include a small test suite plus sample documents that prove accuracy on typical use cases.