TAG (Table-Augmented Generation)

SELECT r.region_cd
FROM region r
LEFT JOIN rate_header h
  ON h.region_cd = r.region_cd
 AND h.release_id = :next
WHERE r.active_flg = 'Y'
  AND h.rate_id IS NULL;

TAG goes beyond retrieval or table lookup — it is not a chatbot that just writes SQL. It interprets what the user is trying to achieve, identifies the relevant tables and relationships, builds filtered queries, retrieves and compares data, and helps business teams decide the next data actions.

Business and domain data teams often depend on developers or Scrum teams to investigate tables, understand relationships, write SQL and work out which domain data must change for a production release. Diagnosing data-related issues also consumes development time — especially when it's unclear whether a problem lies in domain data or in application code.

Every investigation routed through developers slows releases and pulls engineering time away from product work. Giving domain data teams direct, safe access to answers helps them investigate data-related issues without always depending on development teams.

I independently architected an application that lets these teams interact with existing domain databases through natural language.

1. 01Natural language request
2. 02Intent understanding
3. 03Domain / entity resolution
4. 04Schema & relationship discovery
5. 05SQL generation
6. 06SQL validation
7. 07Query execution
8. 08Result analysis
9. 09Data action recommendations

• Understands business terminology and user intent
• Maps business concepts to entities, tables, columns and values
• Relationships among domain data tables
• Natural-language interpretation of business requests
• Translation of intent into relevant SQL
• SQL with appropriate filters and WHERE clauses
• Execution of generated queries
• Data comparison for differences and inconsistencies
• Identifying tables/data to update for a future release
• Separating domain data issues from suspected code issues

I individually architected the system — defining how a request moves from intent interpretation through schema and relationship discovery, filtered query generation, execution and comparison, to a recommendation the team can act on.

LLM applications, SQL generation and validation, schema and relationship metadata, and data comparison over enterprise relational data. Examples on this page are sanitized and use sample data; no confidential details are shown.