This course contains the use of artificial intelligence.

Most RAG tutorials stop at "embed some text, drop it in a vector database, and call an LLM." That naive pipeline falls apart the moment it meets a real enterprise corpus: dense tables, scanned figures, exact drug codes, ambiguous questions, and answers that absolutely must be faithful and auditable.

This course is the missing engineering playbook for advanced, production-grade Retrieval-Augmented Generation. You will rebuild a realistic system end to end: HelixRAG, a clinical knowledge assistant for a 4,000-employee health system, where every answer must be grounded, cited, measured, and governed before it can ship. Working hands-on against a sensitive, multimodal corpus, you will move the numbers that matter, faithfulness, context recall, cost per query, and latency.

What you will master: advanced chunking and indexing (semantic, document-aware, contextual retrieval, small-to-big); hybrid search that fuses BM25 and dense vectors with Reciprocal Rank Fusion, then sharpens results with cross-encoder re-ranking; query enhancement including rewriting, decomposition, multi-query fan-out, and HyDE; self-correcting and agentic patterns such as Self-RAG, Corrective RAG (CRAG), and Graph RAG; multimodal retrieval over tables, images, and diagrams; and production hardening with automated evaluation pipelines, cost-per-query optimization, observability and tracing, and PHI/PII data governance for HIPAA and GDPR.

Every technique is taught the way a senior engineer would judge it: by measured retrieval and answer quality, cost, and risk, not toy demos. You will finish with a best-practice checklist and a production-grade reference architecture you can adapt to your own domain.

If you are an AI engineer, ML engineer, or developer ready to move beyond naive RAG, this course gives you the patterns, trade-offs, and evaluation discipline to build retrieval systems you can trust in production.