[
https://issues.apache.org/jira/browse/HIVE-29681?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18091578#comment-18091578
]
Zhihua Deng edited comment on HIVE-29681 at 6/25/26 10:24 AM:
--------------------------------------------------------------
The goal(and sub-tasks) is to make Metastore a dynamic metadata repository and
AI(LLM) friendly, instead of simply relying on keywords, such as exact table
name, to get the metadata stored in Metastore. It doesn't focus on text_to_sql,
but should help text_to_sql to get the schema intended performantly and
precisely, and help other engines as well, such as Impala or Spark.
was (Author: dengzh):
The goal(and sub-tasks) is to make Metastore a dynamic metadata repository and
AI(LLM) friendly, instead of simply relying on keywords, such as exact table
name, to get the metadata stored. It doesn't focus on text_to_sql, but should
help text_to_sql to get the schema intended performantly and precisely, and
help other engines as well, such as Impala or Spark.
> Introduce dynamic table discovery to Metastore
> ----------------------------------------------
>
> Key: HIVE-29681
> URL: https://issues.apache.org/jira/browse/HIVE-29681
> Project: Hive
> Issue Type: Sub-task
> Reporter: Zhihua Deng
> Assignee: Zhihua Deng
> Priority: Major
>
> Today, metadata search in MetaStore is effective when users know exact or
> partial table names, database names, or schema fields. However, it performs
> poorly for natural-language or concept-driven queries such as:
> * “monthly customer churn tables”
> * “finance datasets with invoice amount”
> * “profile tables with email and phone”
> * “recommendation feature tables”
> AI agents increasingly expect Metastore to understand intent, synonyms, and
> business terminology, without changing MetaStore’s role as the source of
> truth. This Jira is trying to solve this gap, will attach the design document
> later.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
