Conversational SQL Agent for Self-Service Business Intelligence

Our project focuses on developing an AI-powered Text-to-SQL Agent for Heineken's APAC Analytics Hub to democratize data access. Currently, business teams face significant bottlenecks due to their heavy reliance on technical teams for data retrieval and report generation. To resolve this, the proposed system enables non-technical stakeholders to query key performance indicators (KPIs) from the regional Data Mart using Natural Language.

The AI agent dynamically translates conversational requests in English and Vietnamese into precise SQL queries. It utilizes a tiered Large Language Model architecture, using GPT-4o Mini for rapid intent classification and GPT-4o for complex query generation. To guarantee accuracy and security, the system integrates a Semantic Layer with hardcoded FMCG business rules, an automated self-correction loop to repair invalid SQL, and strict Row-Level Security to enforce regional access control.

Ultimately, this automated workflow eliminates traditional reporting delays and significantly reduces IT dependencies. It empowers Heineken's business teams to respond to Fast-Moving Consumer Goods (FMCG) market dynamics with unprecedented speed and agility.