- Understand the real components of a RAG system.
- Distinguish a slick demo from a system usable at an SMB.
- Define quality criteria before indexing documents.
The complete pipeline
Documentos -> ingesta y limpieza -> chunking -> embeddings -> base vectorial -> recuperación -> reranking -> generación con citas -> evaluación y logs
Design questions
- Which documents can each user query?
- Does every answer need an exact citation?
- What happens if it doesn't find evidence?
- How are outdated documents updated?
- Which inputs might contain malicious instructions?