[
https://issues.apache.org/jira/browse/SOLR-17117?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Suleyman Ekmekci updated SOLR-17117:
------------------------------------
Priority: Minor (was: Major)
> Proposing to add Generative Search (RAG) to Solr
> ------------------------------------------------
>
> Key: SOLR-17117
> URL: https://issues.apache.org/jira/browse/SOLR-17117
> Project: Solr
> Issue Type: New Feature
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: Suleyman Ekmekci
> Priority: Minor
> Labels: Generative-Search, RAG
>
> Retrieval Augmented Generation (RAG) is the technique of optimizing the
> output of a Large Language Model (LLM), so it references an authoritative
> knowledge base outside of its training data sources before generating a
> response. RAG has recently become critical for enhancing the accuracy and
> reliability of generative AI models.
>
> Currently, RAG isn't supported natively by Solr. There might be different
> approaches to make this happen. I had the opportunity to experiment
> generative search with the Solr streaming expressions. Logic will be simple:
> as an input, Solr results and user prompt will be given to a configured LLM.
> LLM will receive the results and based on the prompt and the results, it will
> generate human-like response. Giving the search results from Solr will help
> LLM to not hallucinate.
>
> Here is one example:
> - Solr query: filtering by director (Christopher Nolan)
> - Receiving Solr response: Titles and summaries of relevant movies
> - LLM Interaction: "Write a comprehensive overview of Christopher Nolan's
> filmmaking style, referencing his films"
> - Expected outcome: Summary, comparisons and film details
>
> Another example:
> - Solr query: documentaries about space exploration
> - Receiving Solr response: Titles and summaries of relevant documentaries
> - LLM Interaction: "Create a viewer's guide that highlights the key themes
> and facts presented in space exploration documentaries. Include a brief
> description and critical analysis for each"
> - Expected outcome: A well-structured guide that's both informative and
> engaging, closely tied to the data provided by Solr
>
> I would like to hear about your opinions, ideas and possible solutions to
> discuss about the feature.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
