MartDB generative AI data agent (GPT-4o)
KT
A GPT-4o PoC for KT that answers natural-language questions over the PostgreSQL MartDB with RAG and text-to-SQL, cutting query turnaround from hours to seconds.
- Problem
- Business stakeholders could not query MartDB without SQL, and analysts spent hours on manual queries.
- Design
- Connected PostgreSQL MartDB to an LLM on Azure AI. RAG plus text-to-SQL answers natural-language questions grounded in MartDB. Streamlit PoC interface.
- Rationale
- Chose text-to-SQL over pre-aggregation to keep answers current with the operational mart, and grounded generation with RAG to reduce hallucination on schema terms.
- Result
- Validated LLM querying over the operational mart. Query turnaround from hours to seconds.
- Stack
- Azure AI Studio, GPT-4o, PostgreSQL, Streamlit, Python
Related projects
- PPAS backup & recovery / data lakehouse
- Real-time analytics platform enhancement
- Airflow DAG batch scheduling
- Insurance system migration from on-premises to AWS
- iOS/Android QA automation
- Document platform PoC series — the predecessors of localization.md
- GitHub single-file operation-line challenge