[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:production-ocr-course":3},"\u003Cdiv align=\"center\">\n  \u003Ch1>📄 Production-Grade SLM-Powered OCR Course 📄\u003C\u002Fh1>\n  \u003Ch3>Build a self-scaling, event-driven OCR pipeline on Kubernetes (AKS \u002F GKE) with Qwen 3.5 + the GLM-OCR SDK\u003C\u002Fh3>\n\u003C\u002Fdiv>\u003Cbr \u002F>\u003Cp align=\"center\">\n    \u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002Fneural-maze\u002Fproduction-ocr-course\u002FHEAD\u002Fimages\u002Farch_overview.jpeg\" alt=\"Architecture\" width=\"700\" \u002F>\n\u003C\u002Fp>\u003Ch2>Table of Contents\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"#table-of-contents\" rel=\"nofollow ugc noopener\">Table of Contents\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#course-overview\" rel=\"nofollow ugc noopener\">Course Overview\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#who-is-this-course-for\" rel=\"nofollow ugc noopener\">Who is this course for?\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#course-breakdown-week-by-week\" rel=\"nofollow ugc noopener\">Course Breakdown: Week by Week\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#getting-started\" rel=\"nofollow ugc noopener\">Getting Started\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#-beyond-traditional-ocr-the-slm-advantage\" rel=\"nofollow ugc noopener\">✨ Beyond Traditional OCR: The SLM Advantage\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#-pipeline-architecture-deep-dive-into-throughput--scaling\" rel=\"nofollow ugc noopener\">🧠 Pipeline Architecture: Deep Dive into Throughput & Scaling\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#-the-exact-document-workflow\" rel=\"nofollow ugc noopener\">🔄 The Exact Document Workflow\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#-robustness-why-a-pre-layout-encoder-improves-fidelity\" rel=\"nofollow ugc noopener\">🧩 Robustness: Why a Pre-layout Encoder Improves Fidelity\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#-technical-report-document-handoff\" rel=\"nofollow ugc noopener\">📑 Technical Report: Document Handoff\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#%EF%B8%8F-formal-architecture-assessment-production-robustness\" rel=\"nofollow ugc noopener\">🏛️ Formal Architecture Assessment: Production Robustness\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#-scaling-philosophy-real-time-workloads-metric-driven\" rel=\"nofollow ugc noopener\">📈 Scaling Philosophy: Real-Time Workloads (Metric-Driven)\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#the-tech-stack\" rel=\"nofollow ugc noopener\">The tech stack\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#contributors\" rel=\"nofollow ugc noopener\">Contributors\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"#license\" rel=\"nofollow ugc noopener\">License\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>Course Overview\u003C\u002Fh2>\n\u003Cp>Most OCR tutorials stop at \"call an API and get some text back.\" This isn't that.\u003C\u002Fp>\n\u003Cp>Instead, we're building a \u003Cstrong>production-grade, self-scaling Visual Document Understanding pipeline\u003C\u002Fstrong>, deployed for real on Kubernetes (AKS or GKE), that goes far beyond flat text extraction: it reasons about charts, tables, and layout the way a human reader would — powered by a Small Language Model (\u003Cstrong>Qwen 3.5\u003C\u002Fstrong>) instead of a bloated frontier model.\u003C\u002Fp>\n\u003Cp>By the end of this course, you'll have your own event-driven OCR system capable of:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>🧠 Understanding documents, not just transcribing them — charts, tables, handwriting, and contextual reasoning via \u003Cstrong>Qwen 3.5 (4B)\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>⚡ Serving generative OCR at \u003Cstrong>1.86 pages\u002Fsecond\u003C\u002Fstrong> with vLLM's continuous batching, PagedAttention, and Multi-Token Prediction (MTP)\u003C\u002Fli>\n\u003Cli>🦀 Ingesting files through a high-concurrency \u003Cstrong>Rust (Axum) gateway\u003C\u002Fstrong>, decoupled from GPU inference via Redis\u003C\u002Fli>\n\u003Cli>🔄 Running a \u003Cstrong>zero-copy, \u003Ccode>\u002Fdev\u002Fshm\u003C\u002Fcode>-based\u003C\u002Fstrong> document handoff between the layout encoder and the inference engine\u003C\u002Fli>\n\u003Cli>☸️ Auto-scaling T4 (layout) and A100 (inference) node pools independently with \u003Cstrong>KEDA\u003C\u002Fstrong>, from zero to bursting load\u003C\u002Fli>\n\u003Cli>🔒 Locking the whole pipeline behind an \u003Cstrong>Internal Load Balancer + Enterprise API Gateway\u003C\u002Fstrong> (Azure APIM \u002F GCP API Gateway), with zero public exposure\u003C\u002Fli>\n\u003Cli>🤖 Wrapping the pipeline as an \u003Cstrong>MCP server\u003C\u002Fstrong> for native use by AI agents, including Claude Code\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Excited? Let's get started!\u003C\u002Fp>\n\u003Chr \u002F>\n\u003Ctable>\n  \u003Ctr>\n    \u003Ctd>\n      \u003Ca href=\"https:\u002F\u002Ftheneuralmaze.substack.com\u002F\" rel=\"nofollow ugc noopener\">\n        \u003Cimg src=\"https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F151655127?s=400&u=2fff53e8c195ac155e5c8ee65c6ba683a72e655f&v=4\" alt=\"The Neural Maze Logo\" width=\"150\" \u002F>\n      \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd>\n      \u003Cdiv>\n        \u003Ch2>📬 Stay Updated\u003C\u002Fh2>\n        \u003Cp>\u003Cb>\u003Ca href=\"https:\u002F\u002Ftheneuralmaze.substack.com\u002F\" rel=\"nofollow ugc noopener\">Join The Neural Maze\u003C\u002Fa>\u003C\u002Fb> and learn to build AI Systems that actually work, from principles to production. Every Wednesday, directly to your inbox. Don't miss out!\u003C\u002Fp>\n      \u003C\u002Fdiv>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Ftheneuralmaze.substack.com\u002F\" rel=\"nofollow ugc noopener\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fstatic\u002Fv1?label&logo=substack&message=Subscribe%20Now&style=for-the-badge&color=black&scale=2\" alt=\"Subscribe Now\" height=\"40\" \u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\u003Chr \u002F>\n\u003Ch2>Who is this course for?\u003C\u002Fh2>\n\u003Cp>This course is for ML\u002FAI Engineers and Platform Engineers who already know how to call an OCR API and want to know **what it takes to run one in \u003C\u002Fp>\n",1785125872947]