Username: [[ username ]]

Librarian

Librarian is a RAG application built by the Developer Programs Solutions Engineering team.

Using hybrid search (a mixture of semantic and keyword techniques), it pulls from a variety of NVIDIA asset types for the information most relevant to your query. Librarian then sends those sources to an LLM to generate a summary of the retrieved content.

Guide: Advanced Search Techniques

Double-quoting keywords: Users can use double quotes to explicitly signal keywords or keyphrases they want to include when filtering search results. This is helpful if you suspect a keyword may not be recognized automatically. For example, will ensure every result includes the keyword term rapids. Without the quotes, there is the small possibility of the classifier system interpreting this as a semantic search with a similar meaning to fast download.

Wildcard keywords: Use double-quotes and an asterisk symbol for filtering on keywords where you have a partial string. For example, will match on both keywords H200 and GH200.

Fuzzy matching: Do not use double-quotes for fuzzy matching. Instead, surround individual fuzzy match words with percent symbols. Use a single percent symbol for the most strict fuzzy match, two percent symbols for a medium fuzzy match, and three percent symbols for the loosest fuzzy match. Do not include spaces or punctuation between your percent symbols. For example, has a strict fuzzy match on Initialisation and a loose fuzzy match on Scario.

Adding a year: If you specifically want content that came out in a particular year (like the current year), include that year at the end of your search. For example, will grab Omniverse-related content from GTC 2024. If a search query leaves out the year but seems to have the intent of asking for a recent event or announcement, the system will try to add the year as a search term under-the-hood. For example,

I want to learn more about...
Your search results contained these related topics...
LLM (for Summarization and Classification)

Asset Types

[[ atype_group['group'] ]]

System Log

(A glimpse into the internals of the RAG application.)

Share or bookmark your search for later reference.

Copy all recommendations to clipboard.

Response

(This text was generated by an LLM.)

Copied to clipboard.
Feedback submitted.