Projects

This assistant uses retrieval-augmented generation (RAG): before answering, it searches the SURF documentation and knowledge base, then bases its reply on what it finds. Each question goes through these steps:

1. Understands your question Follow-ups like “and for SHYFEM?” are rewritten into a complete, standalone search using the conversation so far.
2. Searches two ways at once It combines keyword matching (exact terms like namelists or version numbers) with meaning-based search, so relevant passages are found even when your wording differs.
3. Ranks the best matches The top candidates are re-scored for how well they actually answer your question, and only the strongest are kept.
4. Answers from the sources Those passages are given to a Large Language Model, which writes the reply grounded in them — not on general knowledge alone.

When sources are used they appear as numbered citations (e.g. [1]) — click any one to open that source beside the chat.