This is RAG — the architecture behind production AI that answers from your documents instead of guessing. Its knowledge here: the real 2026 World Cup, all 48 teams.
The constellation above is the corpus (fancy Latin word for "body of documents"). Teams still in the cup orbit the core; the eliminated drift to the rim — updated daily. Ask, and watch the teams in the answer light up.
RAG is how production AI answers from your data instead of guessing. Four layers did the work — here's each one.
The knowledge the system can draw on. Here it's the 2026 World Cup — teams, groups, and results. Swap this seam and the same demo answers from a client's documents.
Every entry is turned into a vector so the system can find what's relevant to your question by meaning, not keywords. That retrieval step is what lights up the graph above.
The retrieved facts are handed to a frontier language model, which writes a grounded answer — anchored in the corpus, not invented.
All four layers sit behind a reusable framework with a swappable corpus and a pluggable embedding provider. This page is its first consumer — not a one-off.