Retrieval-Augmented Generation and Knowledge Systems

About this course

Retrieval-augmented generation is how most enterprise LLM systems reach their own data. This course builds the full pipeline to production quality: ingestion and chunking, embedding models and vector indexes, hybrid retrieval and reranking, query understanding, knowledge graphs and structured and multimodal retrieval. It treats evaluation as the core discipline, with metrics that locate failures stage by stage, and finishes with the access-control, privacy, freshness and cost concerns that decide whether a RAG system can be trusted with enterprise documents.

For: software engineer, ml practitioner, data professional, architect.

Learning outcomes

On successful completion you will be able to:

1. 1Explain the retrieval-augmented generation pipeline (ingestion, chunking, embedding, indexing, retrieval, reranking and generation) and the characteristic failures of each stage.
2. 2Build a production-grade RAG service with hybrid retrieval, reranking, metadata filtering and grounded, cited answers.
3. 3Analyse retrieval and answer quality with appropriate metrics, and diagnose whether a failure originates in parsing, chunking, retrieval, ranking or generation.
4. 4Evaluate alternative knowledge architectures (vector, hybrid, graph, structured and multimodal retrieval, or long context instead of retrieval) for a given corpus and query mix.
5. 5Design an evaluation harness for a RAG system that combines retrieval metrics, LLM-as-judge scoring and human review.
6. 6Take responsibility for access control, personal-data protection and freshness in a RAG system that handles sensitive enterprise documents.

Modules and lessons

1. Week 1RAG fundamentals and ingestion⌄

   • Why retrieval: the RAG architecture
   • Document ingestion, parsing and chunkinglab
2. Week 2Embeddings and vector search⌄

   • Embedding models for retrieval
   • Vector indexes and databaseslab
3. Week 3Retrieval quality⌄

   • Hybrid retrieval and reranking
   • Query understanding and transformationlab
4. Week 4Beyond text chunks⌄

   • Knowledge graphs and GraphRAGcase study
   • Structured and multimodal retrieval
5. Week 5Evaluating RAG⌄

   • Measuring retrieval and generation qualitylab
   • Diagnosing and improving a RAG system
6. Week 6RAG in production⌄

   • Access control, privacy and data protection in RAG
   • Operating RAG at scale

Assessment

• Quiz (20%)

  Six weekly quizzes drawn from the lesson quiz banks; best five count.

• Lab (30%)

  Four graded labs that build the pipeline stage by stage, each with a measured before-and-after result.

• Project (35%)

  Build and evaluate a RAG service over a provided document corpus with per-user permissions; deliver code, an evaluation report and a one-page design note.

• Design review (15%)

  Mentor review of the design note, covering the retrieval architecture chosen, the alternatives rejected and the access-control and privacy design.

Pass mark 60%. Grades: Distinction, Merit, Pass.