Module 04: RAG Engineering

Goal

Build retrieval-augmented generation systems that produce grounded answers with citations, access control, evaluation, and operational visibility.

Core Architecture

Documents -> Parsing -> Cleaning -> Chunking -> Embeddings -> Vector Store -> Retrieval -> Reranking -> Context Builder -> LLM -> Answer + Citations

Core Topics

• Document parsing.
• Cleaning and normalization.
• Chunking strategies.
• Semantic chunking.
• Metadata design.
• Embeddings.
• Vector search.
• Hybrid search.
• Keyword search.
• Reranking.
• Query rewriting.
• Context compression.
• Retrieval evaluation.
• Answer evaluation.
• Hallucination detection.

Study Tasks

• Parse PDF, DOCX, spreadsheet, HTML, and plain text documents.
• Compare fixed-size chunking, heading-based chunking, and semantic chunking.
• Create metadata filters.
• Implement vector search.
• Implement hybrid search.
• Add citations.
• Create an evaluation dataset.
• Measure retrieval quality and answer faithfulness.

Project

Build projects/04-enterprise-rag-platform/.

Tips

• Bad chunking creates bad answers.
• Store source metadata from the beginning.
• Retrieval quality should be measured before answer quality.
• Keep access control outside the prompt; enforce it in retrieval and APIs.

Recommended Tools

• PostgreSQL with pgvector.
• Qdrant, Weaviate, Pinecone, Redis, or Elasticsearch/OpenSearch.
• LlamaIndex or LangChain for comparison, not dependency lock-in.

Completion Checklist

• Can explain RAG vs fine-tuning.
• Can design chunking and metadata strategy.
• Can evaluate retrieval precision and recall.
• Can produce citation-backed answers.
• Can protect restricted documents.