{"repo":"MichaelWalker-git/deepseek_ocr","free":true,"listed":false,"github":"https://github.com/MichaelWalker-git/deepseek_ocr","clone":"git clone https://github.com/MichaelWalker-git/deepseek_ocr.git","description":"Self hosting your own DeepSeek OCR model in AWS","language":"Python","stars":11,"topics":["deepseek-ocr","llm","self-hosted"],"license":"MIT-0","category":"self-hosted-apps","readme_excerpt":"DeepSeek OCR Pipeline - CDK Implementation A production-ready, 100% accuracy OCR pipeline using DeepSeek-OCR with AWS CDK, ECS, and A2I human review workflows. 🏗️ Architecture Overview This project implements a hybrid architecture that combines the proven Bogdanovich77 DeepSeek-OCR Docker implementation with enterprise-grade AWS orchestration to achieve 100% accuracy through human-in-the-loop validation. Key Architecture Decisions Why Hybrid Approach? - ✅ Proven OCR Solution : Leverages the battle-tested Bogdanovich77 Docker implementation - ✅ Enterprise Orchestration : AWS Step Functions + A2I for workflow management - ✅ Cost Optimization : 60% cost reduction vs pure SageMaker approach - ✅ Predictable Performance : No cold starts, model stays loaded in memory - ✅ Scalable : Auto-scaling from 1-10 GPU instances based on demand Technology Stack: - Container Runtime : ECS on EC2 with g4dn.xlarge GPU instances - Model Storage : Baked into Docker image ( 15GB) for faster scaling - API Layer : API Gateway with VPC integration - Human Review : Amazon A2I with MTurk workforce - Data Storage : S3 + DynamoDB with intelligent lifecycle policies - Orchestration : Step Functions for end-to-end workflow 📋 System Requirements Hardware Requirements - GPU Instances : g4dn.xlarge (1x NVIDIA T4 GPU, 16GB VRAM) - Auto Scaling : 1-10 instances based on CPU, memory, and request count - Storage : 100GB GP3 EBS per instance Software Dependencies - AWS CDK : v2.221.0+ - Node.js : v18+ - Docker : F","default_branch":null,"files":null,"tree":[],"storefront":"/r/MichaelWalker-git","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/MichaelWalker-git/deepseek_ocr/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}