Reference data
This is the knowledge base a GenAI assistant can access.
The data is stored in a refinery project (open source product), but it is not automatically created when a new project is created or imported. The corresponding refinery project can be used to e.g. RAG (Retrieval Augmented Generation) use cases (most common use case to include refinery in Cognition), monitor PII in your data, quality of your knowledge database and much more.
The data from the refinery project can be mapped to the cognition project using one of the strategy steps: ‘Neural search’. The configuration of the strategy plays huge role in this setup. In the Neural Search strategy node, few configurations needs to be setup such as: target refinery project, embedding, include score or not and most importantly choose map attribute to reference. The attribute ‘reference’ is expected but even if it’s not present in the dataset, there is still the option to automatically convert the name.
Example configuration of the Neural Search Node:
The possibilities that refinery provides are described in details in the documentation for refinery: https://docs.kern.ai/.