📄️ RAG systems

Build a retrieval-augmented generation pipeline over 300 internal procedures, from PDF extraction to citations, cache and cost control, with a 40-question exam and a verifiable certificate.

📄️ 1. Why ground in documents

Module 1 of the RAG premium course: what a language model does not know, why RAG beats fine-tuning for private and moving content, and the shape of the pipeline the rest of the course will build.

📄️ 2. Extracting documents

Module 2 of the RAG premium course: reading order, tables, running headers, OCR for scans, and the metadata every extracted chunk must carry to enable citations later on.

📄️ 3. Chunking passages

Module 3 of the RAG premium course: fixed, sentence, section and hierarchical chunking, size in tokens, overlap, parent context, and the table-splitting failure compared on the internal procedures.

📄️ 4. Embeddings and index

Module 4 of the RAG premium course: choosing an embedding model, HNSW indexing in Chroma and pgvector, metadata filters, incremental updates, and the classic forgotten-filter failure.

📄️ 5. Hybrid search

Module 5 of the RAG premium course: BM25 vs vector similarity, reciprocal-rank fusion for hybrid search, short queries and acronyms, and rewriting the question before retrieval.

📄️ 6. Reranking

Module 6 of the RAG premium course: cross-encoders vs bi-encoders, how many candidates to keep after reranking, latency and cost, source diversity, and reranking as a knob you tune.

📄️ 7. Prompt and citations

Module 7 of the RAG premium course: assembling passages into the prompt, ordering the context, forcing abstention, producing verifiable citations, and handling contradictions between sources.

📄️ 8. Evaluation

Module 8 of the RAG premium course: retrieval recall, answer faithfulness, LLM-as-judge, an annotated test set on the red thread, and RAGAS in a glance.

📄️ 9. Caching and cost

Module 9 of the RAG premium course: caching embeddings and answers, per-question cost breakdown, incremental reindexing, and structured logging that survives audits.

📄️ 10. Project

Module 10 of the RAG premium course: wire extraction, chunking, index, hybrid retrieval, rerank, prompt, cache and evaluation into a working document assistant with per-user permissions.

📄️ 11. Recap and exam

Complete recap of the RAG premium course: extraction, chunking, index, hybrid search, reranking, prompt, evaluation, cost, then the 40-question exam and diagnostic tree for wrong answers.