Deckard collects and exposes local knowledge by leveraging modern LLMs. It uses a hybrid Retrieval-Augmented Generation (RAG) approach to process queries and generate accurate, contextually relevant responses.
Some stub documentation is currently available in the documentation folder. Notable topics:
- For information on how queries are processed, see the RAG Query Processing document.
- For details on how the RAG pipeline is built, see the RAG Data Building Process document.
- Python 3.8+
- Poetry
- Nvidia drivers
- CUDA
Start the API server:
poetry run api:start
Build the RAG pipeline's underlying data to ready it for use. This may have requirements such as network requests, database tunnels, or on-disk data files.
poetry run build:rag <pipeline>
poetry run build:rag libpages
Other commands are available for convenience:
Query the configured RAG pipeline:
poetry run query:rag <pipeline> <query>
poetry run query:rag libpages 'Where is the bathroom?'
poetry run query:rag libpages 'Who is the dean of UNB libraries?'
Query the LLM model directly without a rag pipeline:
poetry run query:llm <query> <optional context>
poetry run query:llm 'Who is Donald Sutherland?'
poetry run query:llm 'Who is Donald Sutherland?' 'Donald sutherland is a duck'
Start the slackbot configured locally. Slack auth tokens need to be in ENV:
poetry run slackbot:start
- In line with our 'open' ethos, UNB Libraries makes its applications and workflows freely available to everyone whenever possible.
- As a result, the contents of this repository [unb-libraries/deckard] are licensed under the MIT License. This license explicitly excludes:
- Any content that remains the exclusive property of its author(s).
- The UNB logo and any associated visual identity assets remain the exclusive property of the University of New Brunswick.