RAG from Scratch: Complete Retrieval-Augmented Generation

Descrizione del corso

A hands-on deep dive into Retrieval-Augmented Generation. You will implement every component from scratch: document loading and chunking (fixed-size, semantic, recursive), embedding model selection and fine-tuning, vector database integration (Chroma, FAISS, Pinecone), hybrid BM25+dense retrieval, metadata filtering, re-ranking with cross-encoders, query transformation (HyDE, multi-query, step-back), and answer generation with faithfulness grounding. Covers both naive RAG and advanced patterns like CRAG, self-RAG, and adaptive RAG.

Cosa imparerai

• Curated from public resources. Build RAG pipelines from zero: chunking strategies, embedding models, vector stores, hybrid search, and advanced retrieval patterns.

Programma del corso

1. Why RAG? LLM limitations and knowledge cutoffs (18 min)
2. RAG pipeline architecture overview (22 min)
3. Document loading: PDFs, HTML, markdown, databases (25 min)
4. Fixed-size vs semantic vs recursive chunking (28 min)
5. Embedding models: OpenAI, Cohere, open-source comparison (32 min)
6. Multi-vector retrieval and late interaction models (26 min)
7. Hybrid search: BM25 + dense vectors (30 min)
8. Re-ranking with cross-encoders and Cohere Rerank (27 min)
9. Query transformation: HyDE, multi-query, step-back (35 min)
10. CRAG, self-RAG, and adaptive RAG patterns (40 min)

Struttura del corso

3 sezioni, 10 lezioni, 12h 0m di durata complessiva. Il corso si segue interamente online, al proprio ritmo.

Domande frequenti