Retrieval Strategies for a better RAG Output

Nuclia supports a range of Retrieval-Augmented Generation strategies to help you tailor the AI’s context and improve the quality of its outputs. Starting with the Default Context, you have a solid baseline of relevant information for your model. You can then enhance depth and clarity by Including Textual Hierarchy, ensuring paragraphs are supported by resource titles and summaries.

For scenarios where a single paragraph isn’t enough, Neighbouring Paragraphs expand the context by incorporating adjacent text. If your questions are complex or your data too scattered, leverage Pre-Queries to gather additional information before the main query, and use Prefilter Pre-Queries to refine the scope of results before searching deeper.

When more comprehensive data is needed, you can Pass Entire Resources as Context, delivering full documents rather than isolated sections. To add further richness and insights, Add Metadata or Pass Specific Fields as Context, ensuring crucial details from multiple sources are available.

For visual models, Include Images from the page or the paragraph itself to provide richer, multimedia content. If your existing knowledge base doesn’t cover every angle, simply Add Extra Context in the form of custom snippets. Finally, when you know exactly where the right information resides, you can Ask a Specific Resource, bypassing the broader RAG process and zeroing in on the data you need.

By mixing and matching these strategies, you can shape the generative model’s perspective, deliver more accurate responses, and create deeply informative, context-rich experiences.