RAG as Context Engineering

Lecture 8.1

• The problem: context windows are finite, knowledge isn't
• RAG defined: retrieve external information, inject into context, generate
• Context engineering applied to knowledge — not model modification
• The contrast with fine-tuning: fine-tuning changes model weights; RAG changes what's in the context. Same underlying model.
• The pipeline at a glance: get information → put it in context → generate
• Important framing: RAG ≠ vector databases. Vectors are one common implementation; the broader concept is retrieval, returned to in 8.3.

Additional Resources

• Lecture slides