StackMap
Subscribe

ai-engineering-hub vs production-ocr-course

36k-star hub of runnable AI engineering tutorials — LLMs, RAG and agent apps as self-contained projects, including build-code-harness: a Claude-Code-style coding harness rebuilt on CrewAI + E2B. — versus — Runnable course: a production OCR pipeline on Kubernetes — Rust ingestion, Qwen 3.5 (4B) served by vLLM at 1.86 pages/s, Redis queues, KEDA autoscaling. Deploy it on AKS/GKE, not a notebook.

The curated verdict

Same learn-by-deploying shape: ai-engineering-hub is a wide catalog of runnable AI projects; this is one vertical done to production depth — breadth vs one deep dive.

ai-engineering-hubproduction-ocr-course
Stars37k183
Forks6.1k50
LanguageJupyter NotebookRust
LicenseMITApache-2.0
Last activity12 days ago3 days ago
Topicscodingocr, training
Curated connections22

ai-engineering-hub — the curator's take

A tutorial hub, not a tool — each folder is a complete, runnable project with real dependencies. The standout is build-code-harness: it rebuilds a coding-agent harness (planning, memory, checkpoints, sandbox, human gate) one layer at a time on CrewAI + E2B, the best way to understand what harnesses like Claude Code actually do. Content repo caveat: quality varies by folder, and there's a newsletter funnel attached.

production-ocr-course — the curator's take

The rare OCR tutorial that ends in infrastructure instead of an API call: event-driven ingestion in Rust, an SLM (Qwen 3.5 4B) on vLLM with continuous batching, and metric-driven KEDA autoscaling — the whole thing deployable on your own cluster, with an honest architecture assessment included. Read it to learn how document-understanding systems are actually operated. NOT a library you import: it's a course with a reference implementation (183 stars, young), the model choices will age, and if you just need OCR output today the mapped engines (Chandra, MinerU, olmOCR) get you there without a Kubernetes cluster.