AI Engineer – Connect CRM to Local NVIDIA GPU Server with Docker

We have an existing production CRM hosted on Hostinger and a dedicated NVIDIA GPU server at our office. We are looking for an experienced AI/Backend Engineer to build a secure connection between our CRM and locally hosted AI models running on our NVIDIA server. Target architecture: CRM (Hostinger) → Secure API → NVIDIA Server → Dockerized AI Services → CRM The first use case is creating product drafts from images. Workflow: 1. An employee uploads one or more product images through the CRM. 2. The CRM sends the images/job securely to the NVIDIA server. 3. A local Vision/Multimodal AI model analyzes the images. 4. It extracts available product information such as: - Product name/type - Dimensions - Color - Material - Specifications - Text and information visible in the images 5. Information that cannot be determined reliably must remain empty or be flagged for human review. The AI must not invent missing information. 6. Structured results are returned to the CRM. 7. The CRM creates a Draft Product for employee review before approval. Technical requirements: - Docker / Docker Compose - NVIDIA GPU / CUDA - Python - REST API - Local Vision / Multimodal AI models - Secure communication between the Hostinger CRM and our office NVIDIA server - Background job handling - Logging and error handling - The CRM and AI server must remain separate systems - Architecture must allow additional AI containers/models to be added later Important: We already have a working CRM. We are NOT looking to rebuild it. Development and testing must initially be isolated from production. No direct changes to existing production products during testing. Deliverables: - Working Dockerized AI service on the NVIDIA server - Secure CRM ↔ NVIDIA connection - Working image-to-product-draft workflow - End-to-end test: CRM → NVIDIA → AI → CRM Draft Product - Full source code - Dockerfile / Docker Compose - API documentation - Setup and deployment instructions - Basic logging and failure handling We are not looking to train an AI model from scratch. We want to use suitable existing local/open-source models. When applying, please answer: 1. Have you deployed local LLM or Vision models on NVIDIA GPUs? Give an example. 2. Have you connected an existing production application to a separate local GPU/AI server? 3. How would you securely connect a cloud-hosted CRM to an office NVIDIA server? 4. Which local Vision/Multimodal model would you initially test for extracting structured information from furniture product images, and why? Please include relevant previous work, GitHub repositories, or portfolio examples if available. We are looking for someone with practical experience in local AI deployment, Docker, NVIDIA GPUs and backend integration — not only prompt engineering or cloud AI API experience.